new: gpu package (#224 )

add eofl
sync publih with main
2026-09-22 05:57:51 -05:00 · 2025-04-10 16:24:30 +03:00 · 2025-04-10 16:24:05 +03:00 · 2025-04-10 16:24:05 +03:00 · 2025-04-10 16:24:05 +03:00 · 2025-04-10 16:24:05 +03:00
113 changed files with 16179 additions and 7219 deletions
@@ -0,0 +1,59 @@
+name: Bug
+description: File a bug report
+title: "[Bug]: "
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thanks for taking the time to fill out this bug report!
+  - type: textarea
+    id: what-happened
+    attributes:
+      label: What happened?
+      description: Describe the error you encountered.
+      placeholder: <Description>
+    validations:
+      required: true
+  - type: textarea
+    id: expected
+    attributes:
+      label: What is the expected behaviour?
+      description: Describe the way you expected the code to behave.
+      placeholder: <Description>
+  - type: textarea
+    id: code-snippet
+    attributes:
+      label: A minimal reproducible example
+      description: It would really help us to fix the problem if you could provide a code snippet that reproduces the issue.
+      placeholder: <Code snippet>
+  - type: textarea
+    id: python-version
+    attributes:
+      label: What Python version are you on? e.g. python --version
+      description: Also tell us, what package manager are you using e.g. conda, pip, poetry?
+      placeholder: Python3.10
+    validations:
+      required: true
+  - type: textarea
+    id: version
+    attributes:
+      label: FastEmbed version
+      description: What version of FastEmbed are you running? python -c "import fastembed; print(fastembed.__version__)". If you're not on the latest, please upgrade and see if the problem persists.
+      placeholder: v0.5.1
+    validations:
+      required: true
+  - type: dropdown
+    id: os
+    attributes:
+      label: What os are you seeing the problem on?
+      multiple: true
+      options:
+        - Linux
+        - MacOS
+        - Windows
+  - type: textarea
+    id: logs
+    attributes:
+      label: Relevant stack traces and/or logs
+      description: Please copy and paste any relevant raised exceptions. This will be automatically formatted into code, so no need for backticks.
+      render: shell
@@ -0,0 +1,5 @@
+blank_issues_enabled: true
+contact_links:
+  - name: GitHub Community Support
+    url: https://github.com/qdrant/fastembed/discussions
+    about: Please ask and answer questions here.
@@ -0,0 +1,22 @@
+name: Feature
+description: New functionality request
+title: "[Feature]: "
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thanks for taking the time to fill out this report!
+  - type: textarea
+    id: feature-description
+    attributes:
+      label: What feature would you like to request?
+      description: Please provide the description of the feature you would like to request.
+      placeholder: <Description>
+    validations:
+      required: true
+  - type: textarea
+    id: additional-info
+    attributes:
+      label: Is there any additional information you would like to provide?
+      description: Please provide any additional information that you think might be useful.
+      placeholder: <Info>
@@ -0,0 +1,22 @@
+name: Model
+description: Request a new model
+title: "[Model]: "
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thanks for taking the time to fill out this report!
+  - type: textarea
+    id: model-name
+    attributes:
+      label: Which model would you like to support?
+      description: Please provide the name of the model you would like to see supported.
+      placeholder: Link to the model (e.g. on HuggingFace)
+    validations:
+      required: true
+  - type: textarea
+    id: motivation
+    attributes:
+      label: What are the main advantages of this model?
+      description: Please describe the main advantages of this model comparing to the existing ones and provide links to benchmarks if there are any.
+      placeholder: <Description>
@@ -0,0 +1,19 @@
+### All Submissions:
+
+* [ ] Have you followed the guidelines in our Contributing document?
+* [ ] Have you checked to ensure there aren't other open [Pull Requests](../../../pulls) for the same update/change?
+
+<!-- You can erase any parts of this template not applicable to your Pull Request. -->
+
+### New Feature Submissions:
+
+* [ ] Does your submission pass the existing tests?
+* [ ] Have you added tests for your feature?
+* [ ] Have you installed `pre-commit` with `pip3 install pre-commit` and set up hooks with `pre-commit install`?
+
+### New models submission:
+
+* [ ] Have you added an explanation of why it's important to include this model?
+* [ ] Have you added tests for the new model? Were canonical values for tests computed via the original model?
+* [ ] Have you added the code snippet for how canonical values were computed?
+* [ ] Have you successfully ran tests with your changes locally?
@@ -1,8 +1,8 @@
-name: ci 
+name: ci
 on:
  push:
    branches:
-      - master 
+      - master
      - main
 permissions:
  contents: write
@@ -14,12 +14,12 @@ jobs:
      - uses: actions/setup-python@v4
        with:
          python-version: 3.x
-      - run: echo "cache_id=$(date --utc '+%V')" >> $GITHUB_ENV 
+      - run: echo "cache_id=$(date --utc '+%V')" >> $GITHUB_ENV
      - uses: actions/cache@v3
        with:
          key: mkdocs-material-${{ env.cache_id }}
          path: .cache
          restore-keys: |
            mkdocs-material-
-      - run: pip install mkdocs-material mkdocstrings pillow cairosvg mknotebooks
+      - run: pip install mkdocs-material mkdocstrings==0.27.0 pillow cairosvg mknotebooks
      - run: mkdocs gh-deploy --force
@@ -15,7 +15,6 @@ on:
    tags:
      - 'v*'           # Push events to every version tag

-
 jobs:
  deploy:

@@ -1,9 +1,11 @@
 name: Tests
+run-name: Tests (gpu)

 on:
-  push:
-    branches: [ master, main ]
  pull_request:
+    branches: [ master, main, gpu ]
+  workflow_dispatch:
+  

 env:
  CARGO_TERM_COLOR: always
@@ -14,10 +16,11 @@ jobs:
    strategy:
      matrix:
        python-version:
-          - '3.8.x'
          - '3.9.x'
          - '3.10.x'
          - '3.11.x'
+          - '3.12.x'
+          - '3.13.x'
        os:
          - ubuntu-latest

@@ -26,16 +29,19 @@ jobs:
    name: Python ${{ matrix.python-version }} on ${{ matrix.os }} test

    steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
      - name: Set up Python
-        uses: actions/setup-python@v2
+        uses: actions/setup-python@v5
        with:
          python-version: ${{ matrix.python-version }}
      - name: Install dependencies
        run: |
          python -m pip install poetry
          poetry config virtualenvs.create false
-          poetry install --no-interaction --no-ansi
-      - name: Run tests
-        run: pytest
-        shell: bash
+          poetry install --no-interaction --no-ansi --without dev,docs
+
+      - name: Run pytest
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        run: |
+          poetry run pytest
@@ -0,0 +1,40 @@
+name: type-checkers
+
+on: [push]
+
+jobs:
+  build:
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: true
+      matrix:
+        python-version: ["3.9", "3.10", "3.11", "3.12", "3.13"]
+        os: [ubuntu-latest]
+
+    name: Python ${{ matrix.python-version }} test
+
+    steps:
+      - uses: actions/checkout@v1
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v2
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip poetry
+          poetry install --no-interaction --no-ansi --without dev,docs,test
+
+          poetry run pip install "numpy<2.0.0"  # https://github.com/python/mypy/issues/17396
+
+      - name: mypy
+        run: |
+            poetry run mypy fastembed \
+              --disallow-incomplete-defs \
+              --disallow-untyped-defs \
+              --disable-error-code=import-untyped
+
+      - name: pyright
+        run: |
+          poetry run pyright tests/type_stub.py
@@ -85,28 +85,8 @@ ipython_config.py
 # pyenv
 #   For a library or package, you might want to ignore these files since the code is
 #   intended to run in multiple environments; otherwise, check them in:
-# .python-version
+.python-version

-# pipenv
-#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
-#   However, in case of collaboration, if having platform-specific dependencies or dependencies
-#   having no cross-platform support, pipenv may install dependencies that don't work, or not
-#   install all needed dependencies.
-#Pipfile.lock
-
-# poetry
-#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
-#   This is especially recommended for binary packages to ensure reproducibility, and is more
-#   commonly ignored for libraries.
-#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
-#poetry.lock
-
-# pdm
-#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
-#pdm.lock
-#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
-#   in version control.
-#   https://pdm.fming.dev/#use-with-ide
 .pdm.toml

 # PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
@@ -152,27 +132,11 @@ dmypy.json
 # Cython debug symbols
 cython_debug/

-# PyCharm
-#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
-#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
-#  and can be added to the global gitignore or merged into this file.  For a more nuclear
-#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
-#.idea/
+.idea/
 .DS_Store
-nbs/*.tar.gz
 *.tar.gz
-nbs/fast-*/*
-local_cache/*/*
-*/local_cache/*/*
-*/*/local_cache/*/*
+**/local_cache/
 docs/experimental/*.parquet
 docs/experimental/*.bin
 qdrant_storage/*
-fooling_around/fast-multilingual-e5-large/config.json
-fooling_around/fast-multilingual-e5-large/model_optimized.onnx
-fooling_around/fast-multilingual-e5-large/model_optimized.onnx.data
-fooling_around/fast-multilingual-e5-large/ort_config.json
-fooling_around/fast-multilingual-e5-large/sentencepiece.bpe.model
-fooling_around/fast-multilingual-e5-large/special_tokens_map.json
-fooling_around/fast-multilingual-e5-large/tokenizer_config.json
-fooling_around/fast-multilingual-e5-large/tokenizer.json
+experiments/models/*
@@ -1,12 +1,9 @@
 repos:
- repo: https://github.com/pre-commit/pre-commit-hooks
-  rev: v3.2.0
-  hooks:
-  - id: trailing-whitespace
-  - id: end-of-file-fixer
-  - id: check-yaml
-  - id: check-added-large-files
- repo: https://github.com/psf/black
-  rev: 23.7.0
-  hooks:
-    - id: black
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.3.4
+    hooks:
+      - id: ruff
+        types_or: [ python, pyi, jupyter ]
+        args: [ --fix ]
+      - id: ruff-format
+        types_or: [ python, pyi, jupyter ]
@@ -0,0 +1,78 @@
+# Contributing to FastEmbed!
+
+:+1::tada: First off, thanks for taking the time to contribute! :tada::+1:
+
+The following is a set of guidelines for contributing to FastEmbed. These are mostly guidelines, not rules. Use your best judgment, and feel free to propose changes to this document in a pull request.
+
+## Table Of Contents
+
+[I don't want to read this whole thing, I just have a question!!!](#i-dont-want-to-read-this-whole-thing-i-just-have-a-question)
+
+[How Can I Contribute?](#how-can-i-contribute)
+  * [Your First Code Contribution](#your-first-code-contribution)
+  * [Adding New Models](#adding-new-models)
+
+[Styleguides](#styleguides)
+  * [Code Lint](#code-lint)
+  * [Pre-Commit Hooks](#pre-commit-hooks)
+ 
+## I don't want to read this whole thing I just have a question!!!
+
+> **Note:** Please don't file an issue to ask a question. You'll get faster results by using the resources below:
+
+* [FastEmbed Docs](https://qdrant.github.io/fastembed/)
+* [Qdrant Discord](https://discord.gg/Qy6HCJK9Dc)
+
+## How Can I Contribute?
+
+## How Do I Submit A (Good) Bug Report?
+
+Bugs are tracked as [GitHub issues](https://guides.github.com/features/issues/). 
+
+Explain the problem and include additional details to help maintainers reproduce the problem:
+
+* **Use a clear and descriptive title** for the issue to identify the problem.
+* **Describe the exact steps which reproduce the problem** in as many details as possible. For example, start by explaining how you are using FastEmbed, e.g. with Langchain, Qdrant Client, Llama Index and which command exactly you used. When listing steps, **don't just say what you did, but explain how you did it**.
+* **Provide specific examples to demonstrate the steps**. Include links to files or GitHub projects, or copy/pasteable snippets, which you use in those examples. If you're providing snippets in the issue, use [Markdown code blocks](https://help.github.com/articles/markdown-basics/#multiple-lines).
+* **Describe the behavior you observed after following the steps** and point out what exactly is the problem with that behavior.
+* **Explain which behavior you expected to see instead and why.**
+* **If the problem is related to performance or memory**, include a [call stack profile capture](https://github.com/joerick/pyinstrument) and your observations. 
+
+Include details about your configuration and environment:
+
+* **Which version of FastEmbed are you using?** You can get the exact version by running `python -c "import fastembed; print(fastembed.__version__)"`.
+* **What's the name and version of the OS you're using**?
+* **Which packages do you have installed?** You can get that list by running `pip freeze`
+
+### Your First Code Contribution
+
+Unsure where to begin contributing to FastEmbed? You can start by looking through these `good-first-issue`issues:
+
+* [Good First Issue](https://github.com/qdrant/fastembed/labels/good%20first%20issue) - issues which should only require a few lines of code, and a test or two. These are a great way to get started with FastEmbed. This includes adding new models which are already tested and ready on Huggingface Hub. 
+
+## Pull Requests
+
+The best way to learn about the mechanics of FastEmbed is to start working on it. 
+
+### Your First Code Contribution
+Your first code contribution can be small bug fixes:
+1. This PR adds a small bug fix for a single input: https://github.com/qdrant/fastembed/pull/148
+2. This PR adds a check for the right file location and extension, specific to an OS: https://github.com/qdrant/fastembed/pull/128
+
+Even documentation improvements and tests are most welcome:
+1. This PR fixes a README link: https://github.com/qdrant/fastembed/pull/143
+
+### Adding New Models
+1. Open Requests for New Models are [here](https://github.com/qdrant/fastembed/labels/model%20request). 
+2. There are quite a few pull requests that were merged for this purpose and you can use them as a reference. Here is an example: https://github.com/qdrant/fastembed/pull/129
+3. Make sure to add tests for the new model
+    - The CANONICAL_VECTOR values must come from a reference implementation usually from Huggingface Transformers or Sentence Transformers
+    - Here is a reference [Colab Notebook](https://colab.research.google.com/drive/1tNdV3DsiwsJzu2AXnUnoeF5av1Hp8HF1?usp=sharing) for how we will evaluate whether your VECTOR values in the test are correct or not.
+
+## Styleguides
+
+### Code Lint
+We use ruff for code linting. It should be installed with poetry since it's a dev dependency.
+
+### Pre-Commit Hooks
+We use pre-commit hooks to ensure that the code is linted before it's committed. You can install pre-commit hooks by running `pre-commit install` in the root directory of the project.
@@ -0,0 +1,22 @@
+Copyright 2024 Qdrant
+
+This product includes software developed by Qdrant
+
+This distribution includes the following Jina AI models, each with its respective license:
+- jinaai/jina-colbert-v2
+    - License: cc-by-nc-4.0
+- jinaai/jina-reranker-v2-base-multilingual
+    - License: cc-by-nc-4.0
+- jinaai/jina-embeddings-v3
+    - License: cc-by-nc-4.0
+
+These models are developed by Jina (https://jina.ai/) and are subject to Jina AI's licensing terms.
+
+This distribution includes the following Google models, each with its respective license:
+- vidore/colpali-v1.3
+    - License: gemma
+
+Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms
+
+Additional Notes:
+This project also includes third-party libraries with their respective licenses. Please refer to the documentation of each library for details regarding its usage and licensing terms.
@@ -1,61 +1,281 @@
 # ⚡️ What is FastEmbed?

-FastEmbed is an easy to use -- lightweight, fast, Python library built for retrieval embedding generation. 
+FastEmbed is a lightweight, fast, Python library built for embedding generation. We [support popular text models](https://qdrant.github.io/fastembed/examples/Supported_Models/). Please [open a GitHub issue](https://github.com/qdrant/fastembed/issues/new) if you want us to add a new model.

-The default embedding supports "query" and "passage" prefixes for the input text. The default model is Flag Embedding, which is top of the [MTEB](https://huggingface.co/spaces/mteb/leaderboard) leaderboard. Here is an example for [Retrieval Embedding Generation](https://qdrant.github.io/fastembed/examples/Retrieval%20with%20FastEmbed/)
+The default text embedding (`TextEmbedding`) model is Flag Embedding, presented in the [MTEB](https://huggingface.co/spaces/mteb/leaderboard) leaderboard. It supports "query" and "passage" prefixes for the input text. Here is an example for [Retrieval Embedding Generation](https://qdrant.github.io/fastembed/qdrant/Retrieval_with_FastEmbed/) and how to use [FastEmbed with Qdrant](https://qdrant.github.io/fastembed/qdrant/Usage_With_Qdrant/).

-1. Light
-    - Quantized model weights
-    - ONNX Runtime for inference
-    - No hidden dependencies on PyTorch or TensorFlow via Huggingface Transformers
+## 📈 Why FastEmbed?

-2. Accuracy/Recall
-    - Better than OpenAI Ada-002
-    - Default is Flag Embedding, which is top of the [MTEB](https://huggingface.co/spaces/mteb/leaderboard) leaderboard
+1. Light: FastEmbed is a lightweight library with few external dependencies. We don't require a GPU and don't download GBs of PyTorch dependencies, and instead use the ONNX Runtime. This makes it a great candidate for serverless runtimes like AWS Lambda. 

-3. Fast
-    - About 2x faster than Huggingface (PyTorch) transformers on single queries
-    - Lot faster for batches!
-    - ONNX Runtime allows you to use dedicated runtimes for even higher throughput and lower latency 
+2. Fast: FastEmbed is designed for speed. We use the ONNX Runtime, which is faster than PyTorch. We also use data parallelism for encoding large datasets.
+
+3. Accurate: FastEmbed is better than OpenAI Ada-002. We also [support](https://qdrant.github.io/fastembed/examples/Supported_Models/) an ever-expanding set of models, including a few multilingual models.

 ## 🚀 Installation

-To install the FastEmbed library, pip works: 
+To install the FastEmbed library, pip works best. You can install it with or without GPU support:

 ```bash
 pip install fastembed
+
+# or with GPU support
+
+pip install fastembed-gpu
 ```

-## 📖 Usage
+## 📖 Quickstart

 ```python
-from fastembed.embedding import FlagEmbedding as Embedding
+from fastembed import TextEmbedding

-documents: List[str] = [
-    "passage: Hello, World!",
-    "query: Hello, World!", # these are two different embedding
-    "passage: This is an example passage.",
-    # You can leave out the prefix but it's recommended
-    "fastembed is supported by and maintained by Qdrant." 
+
+# Example list of documents
+documents: list[str] = [
+    "This is built to be faster and lighter than other embedding libraries e.g. Transformers, Sentence-Transformers, etc.",
+    "fastembed is supported by and maintained by Qdrant.",
 ]
-embedding_model = Embedding(model_name="BAAI/bge-base-en", max_length=512) 
-embeddings: List[np.ndarray] = list(embedding_model.embed(documents))
+
+# This will trigger the model download and initialization
+embedding_model = TextEmbedding()
+print("The model BAAI/bge-small-en-v1.5 is ready to use.")
+
+embeddings_generator = embedding_model.embed(documents)  # reminder this is a generator
+embeddings_list = list(embedding_model.embed(documents))
+  # you can also convert the generator to a list, and that to a numpy array
+len(embeddings_list[0]) # Vector of 384 dimensions
 ```

-### Why fast?
+Fastembed supports a variety of models for different tasks and modalities.
+The list of all the available models can be found [here](https://qdrant.github.io/fastembed/examples/Supported_Models/)
+### 🎒 Dense text embeddings

-It's important we justify the "fast" in FastEmbed. FastEmbed is fast because:
+```python
+from fastembed import TextEmbedding

-1. Quantized model weights
-2. ONNX Runtime which allows for fast inference on CPU and other dedicated runtimes
+model = TextEmbedding(model_name="BAAI/bge-small-en-v1.5")
+embeddings = list(model.embed(documents))

-### Why light?
-1. No hidden dependencies on PyTorch or TensorFlow via Huggingface Transformers
-2. We do use the tokenizer from Huggingface Transformers, but it's a light dependency
+# [
+#   array([-0.1115,  0.0097,  0.0052,  0.0195, ...], dtype=float32),
+#   array([-0.1019,  0.0635, -0.0332,  0.0522, ...], dtype=float32)
+# ]

-### Why accurate?
-1. Better than OpenAI Ada-002
-2. Top of the Embedding leaderboards e.g. [MTEB](https://huggingface.co/spaces/mteb/leaderboard)
+```

-#### Similar Work
-Ilyas M. wrote about using [FlagEmbeddings with Optimum](https://twitter.com/IlysMoutawwakil/status/1705215192425288017) over CUDA.
+Dense text embedding can also be extended with models which are not in the list of supported models.
+
+```python
+from fastembed import TextEmbedding
+from fastembed.common.model_description import PoolingType, ModelSource
+
+TextEmbedding.add_custom_model(
+    model="intfloat/multilingual-e5-small",
+    pooling=PoolingType.MEAN,
+    normalization=True,
+    sources=ModelSource(hf="intfloat/multilingual-e5-small"),  # can be used with an `url` to load files from a private storage
+    dim=384,
+    model_file="onnx/model.onnx",  # can be used to load an already supported model with another optimization or quantization, e.g. onnx/model_O4.onnx
+)
+model = TextEmbedding(model_name="intfloat/multilingual-e5-small")
+embeddings = list(model.embed(documents))
+```
+
+
+### 🔱 Sparse text embeddings
+
+* SPLADE++
+
+```python
+from fastembed import SparseTextEmbedding
+
+model = SparseTextEmbedding(model_name="prithivida/Splade_PP_en_v1")
+embeddings = list(model.embed(documents))
+
+# [
+#   SparseEmbedding(indices=[ 17, 123, 919, ... ], values=[0.71, 0.22, 0.39, ...]),
+#   SparseEmbedding(indices=[ 38,  12,  91, ... ], values=[0.11, 0.22, 0.39, ...])
+# ]
+```
+
+<!--
+* BM42 - ([link](ToDo))
+
+```
+from fastembed import SparseTextEmbedding
+
+model = SparseTextEmbedding(model_name="Qdrant/bm42-all-minilm-l6-v2-attentions")
+embeddings = list(model.embed(documents))
+
+# [
+#   SparseEmbedding(indices=[ 17, 123, 919, ... ], values=[0.71, 0.22, 0.39, ...]),
+#   SparseEmbedding(indices=[ 38,  12,  91, ... ], values=[0.11, 0.22, 0.39, ...])
+# ]
+```
+-->
+
+### 🦥 Late interaction models (aka ColBERT)
+
+
+```python
+from fastembed import LateInteractionTextEmbedding
+
+model = LateInteractionTextEmbedding(model_name="colbert-ir/colbertv2.0")
+embeddings = list(model.embed(documents))
+
+# [
+#   array([
+#       [-0.1115,  0.0097,  0.0052,  0.0195, ...],
+#       [-0.1019,  0.0635, -0.0332,  0.0522, ...],
+#   ]),
+#   array([
+#       [-0.9019,  0.0335, -0.0032,  0.0991, ...],
+#       [-0.2115,  0.8097,  0.1052,  0.0195, ...],
+#   ]),  
+# ]
+```
+
+### 🖼️ Image embeddings
+
+```python
+from fastembed import ImageEmbedding
+
+images = [
+    "./path/to/image1.jpg",
+    "./path/to/image2.jpg",
+]
+
+model = ImageEmbedding(model_name="Qdrant/clip-ViT-B-32-vision")
+embeddings = list(model.embed(images))
+
+# [
+#   array([-0.1115,  0.0097,  0.0052,  0.0195, ...], dtype=float32),
+#   array([-0.1019,  0.0635, -0.0332,  0.0522, ...], dtype=float32)
+# ]
+```
+
+### Late interaction multimodal models (ColPali)
+
+```python
+from fastembed import LateInteractionMultimodalEmbedding
+
+doc_images = [
+    "./path/to/qdrant_pdf_doc_1_screenshot.jpg",
+    "./path/to/colpali_pdf_doc_2_screenshot.jpg",
+]
+
+query = "What is Qdrant?"
+
+model = LateInteractionMultimodalEmbedding(model_name="Qdrant/colpali-v1.3-fp16")
+doc_images_embeddings = list(model.embed_image(doc_images))
+# shape (2, 1030, 128)
+# [array([[-0.03353882, -0.02090454, ..., -0.15576172, -0.07678223]], dtype=float32)]
+query_embedding = model.embed_text(query)
+# shape (1, 20, 128)
+# [array([[-0.00218201,  0.14758301, ...,  -0.02207947,  0.16833496]], dtype=float32)]
+```
+
+### 🔄 Rerankers
+```python
+from fastembed.rerank.cross_encoder import TextCrossEncoder
+
+query = "Who is maintaining Qdrant?"
+documents: list[str] = [
+    "This is built to be faster and lighter than other embedding libraries e.g. Transformers, Sentence-Transformers, etc.",
+    "fastembed is supported by and maintained by Qdrant.",
+]
+encoder = TextCrossEncoder(model_name="Xenova/ms-marco-MiniLM-L-6-v2")
+scores = list(encoder.rerank(query, documents))
+
+# [-11.48061752319336, 5.472434997558594]
+```
+
+Text cross encoders can also be extended with models which are not in the list of supported models.
+
+```python
+from fastembed.rerank.cross_encoder import TextCrossEncoder 
+from fastembed.common.model_description import ModelSource
+
+TextCrossEncoder.add_custom_model(
+    model="Xenova/ms-marco-MiniLM-L-4-v2",
+    model_file="onnx/model.onnx",
+    sources=ModelSource(hf="Xenova/ms-marco-MiniLM-L-4-v2"),
+)
+model = TextCrossEncoder(model_name="Xenova/ms-marco-MiniLM-L-4-v2")
+scores = list(model.rerank_pairs(
+    [("What is AI?", "Artificial intelligence is ..."), ("What is ML?", "Machine learning is ..."),]
+))
+```
+
+## ⚡️ FastEmbed on a GPU
+
+FastEmbed supports running on GPU devices.
+It requires installation of the `fastembed-gpu` package.
+
+```bash
+pip install fastembed-gpu
+```
+
+Check our [example](https://qdrant.github.io/fastembed/examples/FastEmbed_GPU/) for detailed instructions, CUDA 12.x support and troubleshooting of the common issues.
+
+```python
+from fastembed import TextEmbedding
+
+embedding_model = TextEmbedding(
+    model_name="BAAI/bge-small-en-v1.5", 
+    providers=["CUDAExecutionProvider"]
+)
+print("The model BAAI/bge-small-en-v1.5 is ready to use on a GPU.")
+
+```
+
+## Usage with Qdrant
+
+Installation with Qdrant Client in Python:
+
+```bash
+pip install qdrant-client[fastembed]
+```
+
+or 
+
+```bash
+pip install qdrant-client[fastembed-gpu]
+```
+
+You might have to use quotes ```pip install 'qdrant-client[fastembed]'``` on zsh.
+
+```python
+from qdrant_client import QdrantClient
+
+# Initialize the client
+client = QdrantClient("localhost", port=6333) # For production
+# client = QdrantClient(":memory:") # For small experiments
+
+# Prepare your documents, metadata, and IDs
+docs = ["Qdrant has Langchain integrations", "Qdrant also has Llama Index integrations"]
+metadata = [
+    {"source": "Langchain-docs"},
+    {"source": "Llama-index-docs"},
+]
+ids = [42, 2]
+
+# If you want to change the model:
+# client.set_model("sentence-transformers/all-MiniLM-L6-v2")
+# List of supported models: https://qdrant.github.io/fastembed/examples/Supported_Models
+
+# Use the new add() instead of upsert()
+# This internally calls embed() of the configured embedding model
+client.add(
+    collection_name="demo_collection",
+    documents=docs,
+    metadata=metadata,
+    ids=ids
+)
+
+search_result = client.query(
+    collection_name="demo_collection",
+    query_text="This is a query document"
+)
+print(search_result)
+```
@@ -0,0 +1,41 @@
+# Releasing FastEmbed
+
+This is a guide how to release `fastembed` and `fastembed-gpu` packages.
+
+## How to
+
+1. Accumulate changes in the `main` branch.
+2. Bump the version in `pyproject.toml`
+
+3. Rebase the `gpu` branch on `main` and resolve conflicts if occurred:
+
+```bash
+git checkout gpu
+git rebase main
+git push -f origin gpu
+```
+
+4. Draft release notes
+5. Checkout to `main` and create a tag, e.g.:
+
+```bash
+git checkout main
+git tag -a v0.1.0 -m "Release v0.1.0"
+```
+
+6. Checkout `gpu` and create a tag, e.g.:
+
+```bash
+git checkout gpu
+git tag -a v0.1.0-gpu -m "Release v0.1.0"
+```
+
+7. Push tags:
+
+```bash
+git push --tags
+```
+
+8. Verify that both packages have been published successfully on PyPI. Try installing them and verify imports.
+9. Create a release on GitHub with the written release notes.
+
@@ -11,51 +11,135 @@
    "\n",
    "## Quick Start\n",
    "\n",
-    "The fastembed package is designed to be easy to use. The main class is the `Embedding` class. It takes a list of strings as input and returns a list of vectors as output. The `Embedding` class is initialized with a model file."
+    "The fastembed package is designed to be easy to use. We'll be using `TextEmbedding` class. It takes a list of strings as input and returns a generator of vectors.\n",
+    "\n",
+    "> 💡 You can learn more about generators from [Python Wiki](https://wiki.python.org/moin/Generators)"
   ]
  },
  {
   "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 1,
   "id": "ada95c6a",
   "metadata": {},
   "outputs": [],
   "source": [
-    "!pip install fastembed --upgrade # Install fastembed"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "ed81d725",
-   "metadata": {},
-   "source": [
-    "Make the necessary imports, initialize the `Embedding` class, and embed your data into vectors:"
+    "!pip install -Uqq fastembed"
   ]
  },
  {
   "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 2,
   "id": "b61c6552",
   "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "890cc3b969354eec8d149d143e301a7a",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Fetching 9 files:   0%|          | 0/9 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "The model BAAI/bge-small-en-v1.5 is ready to use.\n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "384"
+      ]
+     },
+     "execution_count": 2,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
   "source": [
-    "from typing import List\n",
    "import numpy as np\n",
-    "from fastembed.embedding import DefaultEmbedding\n",
+    "\n",
+    "from fastembed import TextEmbedding\n",
+    "\n",
    "\n",
    "# Example list of documents\n",
-    "documents: List[str] = [\n",
-    "    \"Hello, World!\",\n",
-    "    \"This is an example document.\",\n",
+    "documents: list[str] = [\n",
+    "    \"This is built to be faster and lighter than other embedding libraries e.g. Transformers, Sentence-Transformers, etc.\",\n",
    "    \"fastembed is supported by and maintained by Qdrant.\",\n",
    "]\n",
-    "# Initialize the DefaultEmbedding class with the desired parameters\n",
-    "embedding_model = DefaultEmbedding(model_name=\"BAAI/bge-small-en\", max_length=512)\n",
-    "embeddings: List[np.ndarray] = list(\n",
-    "    embedding_model.embed(documents)\n",
-    ")  # notice that we are casting the generator to a list\n",
    "\n",
-    "print(embeddings[0].shape)"
+    "# This will trigger the model download and initialization\n",
+    "embedding_model = TextEmbedding()\n",
+    "print(\"The model BAAI/bge-small-en-v1.5 is ready to use.\")\n",
+    "\n",
+    "embeddings_generator = embedding_model.embed(documents)\n",
+    "embeddings_list = list(embeddings_generator)\n",
+    "len(embeddings_list[0])  # Vector of 384 dimensions"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "d772190b",
+   "metadata": {},
+   "source": [
+    "> 💡 **Why do we use generators?**\n",
+    "> \n",
+    "> We use them to save memory mostly. Instead of loading all the vectors into memory, we can load them one by one. This is useful when you have a large dataset and you don't want to load all the vectors at once."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "id": "8a225cb8",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Document: This is built to be faster and lighter than other embedding libraries e.g. Transformers, Sentence-Transformers, etc.\n",
+      "Vector of type: <class 'numpy.ndarray'> with shape: (384,)\n",
+      "Document: fastembed is supported by and maintained by Qdrant.\n",
+      "Vector of type: <class 'numpy.ndarray'> with shape: (384,)\n"
+     ]
+    }
+   ],
+   "source": [
+    "embeddings_generator = embedding_model.embed(documents)\n",
+    "\n",
+    "for doc, vector in zip(documents, embeddings_generator):\n",
+    "    print(\"Document:\", doc)\n",
+    "    print(f\"Vector of type: {type(vector)} with shape: {vector.shape}\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "id": "769a1be9",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "(2, 384)"
+      ]
+     },
+     "execution_count": 4,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "embeddings_list = np.array(list(embedding_model.embed(documents)))\n",
+    "embeddings_list.shape"
   ]
  },
  {
@@ -63,128 +147,74 @@
   "id": "8c49ae50",
   "metadata": {},
   "source": [
-    "## Let's think step by step"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "92cf4b76",
-   "metadata": {},
-   "source": [
-    "### Setup\n",
-    "\n",
-    "Importing the required classes and modules:"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "c0a6f634",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from typing import List\n",
-    "import numpy as np\n",
-    "from fastembed.embedding import DefaultEmbedding as Embedding"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "3fd03a71",
-   "metadata": {},
-   "source": [
-    "Notice that we are using the DefaultEmbedding -- which is a quantized, state of the Art Flag Embedding model which beats OpenAI's Embedding by a large margin. \n",
-    "\n",
-    "### Prepare your Documents\n",
-    "You can define a list of documents that you'd like to embed. These can be sentences, paragraphs, or even entire documents. \n",
+    "We're using [BAAI/bge-small-en-v1.5](https://huggingface.co/BAAI/bge-small-en-v1.5) a state of the art Flag Embedding model. The model does better than OpenAI text-embedding-ada-002. We've made it even faster by converting it to ONNX format and quantizing the model for you.\n",
    "\n",
    "#### Format of the Document List\n",
+    "\n",
    "1. List of Strings: Your documents must be in a list, and each document must be a string\n",
-    "2. For Retrieval Tasks: If you're working with queries and passages, you can add special labels to them:\n",
+    "2. For Retrieval Tasks with our default: If you're working with queries and passages, you can add special labels to them:\n",
    "- **Queries**: Add \"query:\" at the beginning of each query string\n",
-    "- **Passages**: Add \"passage:\" at the beginning of each passage string"
+    "- **Passages**: Add \"passage:\" at the beginning of each passage string\n",
+    "\n",
+    "## Beyond the default model\n",
+    "\n",
+    "The default model is built for speed and efficiency. If you need a more accurate model, you can use the `TextEmbedding` class to load any model from our list of available models. You can find the list of available models using `TextEmbedding.list_supported_models()`."
   ]
  },
  {
   "cell_type": "code",
-   "execution_count": null,
-   "id": "145a56ce",
+   "execution_count": 5,
+   "id": "2e9c8766",
   "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "9470ec542f3c4400a42452c2489a1abc",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Fetching 8 files:   0%|          | 0/8 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
   "source": [
-    "# Example list of documents\n",
-    "documents: List[str] = [\n",
-    "    \"passage: Hello, World!\",\n",
-    "    \"query: Hello, World!\",  # these are two different embedding\n",
-    "    \"passage: This is an example passage.\",\n",
-    "    # You can leave out the prefix but it's recommended\n",
-    "    \"fastembed is supported by and maintained by Qdrant.\",\n",
-    "]"
+    "multilingual_large_model = TextEmbedding(\"intfloat/multilingual-e5-large\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "id": "a9e70f0e",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "(4, 1024)"
+      ]
+     },
+     "execution_count": 6,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "np.array(\n",
+    "    list(multilingual_large_model.embed([\"Hello, world!\", \"你好世界\", \"¡Hola Mundo!\", \"नमस्ते!\"]))\n",
+    ").shape  # Vector of 1024 dimensions"
   ]
  },
  {
   "cell_type": "markdown",
-   "id": "1cb3cc87",
+   "id": "64fe20ed",
   "metadata": {},
   "source": [
-    "### Load the Embedding Model Weights\n",
-    "Next, initialize the Embedding class with the desired parameters. Here, \"BAAI/bge-small-en\" is the pre-trained model name, and max_length=512 is the maximum token length for each document.\n",
-    "\n",
-    "This will download the model weights, decompress to directory `local_cache` and load them into the Embedding class.\n",
-    "\n",
-    "#### Initialize DefaultEmbedding\n",
-    "\n",
-    "We will initialize Flag Embeddings with the model name and the maximum token length. That is the DefaultEmbedding class with the model name \"BAAI/bge-small-en\" and max_length=512."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "272c8915",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "embedding_model = DefaultEmbedding()"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "5549d501",
-   "metadata": {},
-   "source": [
-    "### Embed your Documents\n",
-    "\n",
-    "Use the embed method of the embedding model to transform the documents into a List of np.array. The method returns a generator, so we cast it to a list to get the embeddings."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "8013eee9",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "embeddings: List[np.ndarray] = list(\n",
-    "    embedding_model.embed(documents)\n",
-    ")  # notice that we are casting the generator to a list"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "e5b5a6ad",
-   "metadata": {},
-   "source": [
-    "You can print the shape of the embeddings to understand their dimensions. Typically, the shape will indicate the number of dimensions in the vector."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "0d8c8e08",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "print(embeddings[0].shape)  # (384,) or similar output"
+    "Next: Checkout how to use FastEmbed with Qdrant for similarity search: [FastEmbed with Qdrant](https://qdrant.github.io/fastembed/examples/Usage_With_Qdrant/)"
   ]
  }
 ],
@@ -204,7 +234,7 @@
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
-   "version": "3.9.17"
+   "version": "3.10.13"
  }
 },
 "nbformat": 4,
@@ -0,0 +1,421 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "d14d29ebd3592ecb",
+   "metadata": {
+    "collapsed": false
+   },
+   "source": [
+    "# Late Interaction Text Embedding Models\n",
+    "\n",
+    "As of version 0.3.0 FastEmbed supports Late Interaction Text Embedding Models and currently available with one of the most popular embedding model of the family - ColBERT.\n",
+    "\n",
+    "## What is a Late Interaction Text Embedding Model?\n",
+    "\n",
+    "Late Interaction Text Embedding Model is a kind of information retrieval model which performs query and documents interactions at the scoring stage.\n",
+    "In order to better understand it, we can compare it to the models without interaction.  \n",
+    "For instance, if you take a sentence-transformer model, compute embeddings for your documents, compute embeddings for your queries, and just compare them by cosine similarity, then you're retrieving points without interaction.\n",
+    "\n",
+    "It is a pretty much easy and straightforward approach, however we might be sacrificing some precision due to its simplicity. It is caused by several facts: \n",
+    "- there is no interaction between queries and documents at the early stage (embedding generation) nor at the late stage (during scoring). \n",
+    "- we are trying to encapsulate all the document information in only one pooled embedding, and obviously, some information might be lost.\n",
+    "\n",
+    "Late Interaction Text Embedding models are trying to address it by computing embeddings for each token in queries and documents, and then finding the most similar ones via model specific operation, e.g. ColBERT (Contextual Late Interaction over BERT) uses MaxSim operation.\n",
+    "With this approach we can have not only a better representation of the documents, but also make queries and documents more aware one of another.\n",
+    "\n",
+    "For more information on ColBERT and MaxSim operation, you can check out [this blogpost](https://jina.ai/news/what-is-colbert-and-late-interaction-and-why-they-matter-in-search/) by Jina AI.\n",
+    "\n",
+    "## ColBERT in FastEmbed\n",
+    "\n",
+    "FastEmbed provides a simple way to use ColBERT model, similar to the ones it has with `TextEmbedding`.\n",
+    " "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "7f1053b17c810be5",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-03T17:20:26.927643Z",
+     "start_time": "2024-06-03T17:20:25.128994Z"
+    },
+    "collapsed": false
+   },
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/joein/work/qdrant/fastembed/venv/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
+      "  from .autonotebook import tqdm as notebook_tqdm\n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "[{'model': 'colbert-ir/colbertv2.0',\n",
+       "  'dim': 128,\n",
+       "  'description': 'Late interaction model',\n",
+       "  'size_in_GB': 0.44,\n",
+       "  'sources': {'hf': 'colbert-ir/colbertv2.0'},\n",
+       "  'model_file': 'model.onnx'}]"
+      ]
+     },
+     "execution_count": 1,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "from fastembed import LateInteractionTextEmbedding\n",
+    "\n",
+    "LateInteractionTextEmbedding.list_supported_models()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "id": "c2c15893df422631",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-03T17:23:35.764183Z",
+     "start_time": "2024-06-03T17:23:21.630277Z"
+    },
+    "collapsed": false
+   },
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Fetching 5 files:   0%|          | 0/5 [00:00<?, ?it/s]\n",
+      "config.json: 100%|██████████| 743/743 [00:00<00:00, 4.56MB/s]\n",
+      "\n",
+      "tokenizer_config.json: 100%|██████████| 405/405 [00:00<00:00, 3.34MB/s]\n",
+      "Fetching 5 files:  20%|██        | 1/5 [00:00<00:01,  3.64it/s]\n",
+      "tokenizer.json:   0%|          | 0.00/466k [00:00<?, ?B/s]\u001b[A\n",
+      "\n",
+      "special_tokens_map.json: 100%|██████████| 112/112 [00:00<00:00, 727kB/s]\n",
+      "\n",
+      "tokenizer.json: 100%|██████████| 466k/466k [00:00<00:00, 1.48MB/s]\u001b[A\n",
+      "\n",
+      "model.onnx:   0%|          | 0.00/436M [00:00<?, ?B/s]\u001b[A\n",
+      "model.onnx:   2%|▏         | 10.5M/436M [00:00<00:34, 12.2MB/s]\u001b[A\n",
+      "model.onnx:   5%|▍         | 21.0M/436M [00:01<00:20, 20.3MB/s]\u001b[A\n",
+      "model.onnx:   7%|▋         | 31.5M/436M [00:01<00:15, 25.7MB/s]\u001b[A\n",
+      "model.onnx:  10%|▉         | 41.9M/436M [00:01<00:13, 29.4MB/s]\u001b[A\n",
+      "model.onnx:  12%|█▏        | 52.4M/436M [00:01<00:12, 31.9MB/s]\u001b[A\n",
+      "model.onnx:  14%|█▍        | 62.9M/436M [00:02<00:11, 33.7MB/s]\u001b[A\n",
+      "model.onnx:  17%|█▋        | 73.4M/436M [00:02<00:10, 34.9MB/s]\u001b[A\n",
+      "model.onnx:  19%|█▉        | 83.9M/436M [00:02<00:09, 35.5MB/s]\u001b[A\n",
+      "model.onnx:  22%|██▏       | 94.4M/436M [00:03<00:09, 36.1MB/s]\u001b[A\n",
+      "model.onnx:  24%|██▍       | 105M/436M [00:03<00:09, 36.6MB/s] \u001b[A\n",
+      "model.onnx:  26%|██▋       | 115M/436M [00:03<00:08, 36.9MB/s]\u001b[A\n",
+      "model.onnx:  29%|██▉       | 126M/436M [00:03<00:08, 37.1MB/s]\u001b[A\n",
+      "model.onnx:  31%|███▏      | 136M/436M [00:04<00:08, 37.3MB/s]\u001b[A\n",
+      "model.onnx:  34%|███▎      | 147M/436M [00:04<00:07, 37.4MB/s]\u001b[A\n",
+      "model.onnx:  36%|███▌      | 157M/436M [00:04<00:07, 37.4MB/s]\u001b[A\n",
+      "model.onnx:  38%|███▊      | 168M/436M [00:05<00:07, 37.5MB/s]\u001b[A\n",
+      "model.onnx:  41%|████      | 178M/436M [00:05<00:06, 37.6MB/s]\u001b[A\n",
+      "model.onnx:  43%|████▎     | 189M/436M [00:05<00:06, 37.6MB/s]\u001b[A\n",
+      "model.onnx:  46%|████▌     | 199M/436M [00:05<00:06, 37.6MB/s]\u001b[A\n",
+      "model.onnx:  48%|████▊     | 210M/436M [00:06<00:06, 37.5MB/s]\u001b[A\n",
+      "model.onnx:  50%|█████     | 220M/436M [00:06<00:05, 37.5MB/s]\u001b[A\n",
+      "model.onnx:  53%|█████▎    | 231M/436M [00:06<00:05, 37.6MB/s]\u001b[A\n",
+      "model.onnx:  55%|█████▌    | 241M/436M [00:06<00:05, 37.6MB/s]\u001b[A\n",
+      "model.onnx:  58%|█████▊    | 252M/436M [00:07<00:04, 37.6MB/s]\u001b[A\n",
+      "model.onnx:  60%|██████    | 262M/436M [00:07<00:04, 37.7MB/s]\u001b[A\n",
+      "model.onnx:  63%|██████▎   | 273M/436M [00:07<00:04, 37.7MB/s]\u001b[A\n",
+      "model.onnx:  65%|██████▍   | 283M/436M [00:08<00:04, 36.0MB/s]\u001b[A\n",
+      "model.onnx:  67%|██████▋   | 294M/436M [00:08<00:03, 36.4MB/s]\u001b[A\n",
+      "model.onnx:  70%|██████▉   | 304M/436M [00:08<00:03, 36.8MB/s]\u001b[A\n",
+      "model.onnx:  72%|███████▏  | 315M/436M [00:08<00:03, 37.0MB/s]\u001b[A\n",
+      "model.onnx:  75%|███████▍  | 325M/436M [00:09<00:02, 37.3MB/s]\u001b[A\n",
+      "model.onnx:  77%|███████▋  | 336M/436M [00:09<00:03, 30.8MB/s]\u001b[A\n",
+      "model.onnx:  79%|███████▉  | 346M/436M [00:10<00:02, 32.6MB/s]\u001b[A\n",
+      "model.onnx:  82%|████████▏ | 357M/436M [00:10<00:02, 33.9MB/s]\u001b[A\n",
+      "model.onnx:  84%|████████▍ | 367M/436M [00:10<00:01, 34.8MB/s]\u001b[A\n",
+      "model.onnx:  87%|████████▋ | 377M/436M [00:10<00:01, 35.7MB/s]\u001b[A\n",
+      "model.onnx:  89%|████████▉ | 388M/436M [00:11<00:01, 36.2MB/s]\u001b[A\n",
+      "model.onnx:  91%|█████████▏| 398M/436M [00:11<00:01, 36.6MB/s]\u001b[A\n",
+      "model.onnx:  94%|█████████▍| 409M/436M [00:11<00:00, 36.9MB/s]\u001b[A\n",
+      "model.onnx:  96%|█████████▌| 419M/436M [00:11<00:00, 37.1MB/s]\u001b[A\n",
+      "model.onnx:  99%|█████████▊| 430M/436M [00:12<00:00, 37.3MB/s]\u001b[A\n",
+      "model.onnx: 100%|██████████| 436M/436M [00:12<00:00, 35.1MB/s]\u001b[A\n",
+      "Fetching 5 files: 100%|██████████| 5/5 [00:13<00:00,  2.68s/it]\n"
+     ]
+    }
+   ],
+   "source": [
+    "embedding_model = LateInteractionTextEmbedding(\"colbert-ir/colbertv2.0\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 16,
+   "id": "e560b5fa7d63bea3",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-03T17:39:33.400876Z",
+     "start_time": "2024-06-03T17:39:33.397431Z"
+    },
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": [
+    "documents = [\n",
+    "    \"ColBERT is a late interaction text embedding model, however, there are also other models such as TwinBERT.\",\n",
+    "    \"On the contrary to the late interaction models, the early interaction models contains interaction steps at embedding generation process\",\n",
+    "]\n",
+    "queries = [\n",
+    "    \"Are there any other late interaction text embedding models except ColBERT?\",\n",
+    "    \"What is the difference between late interaction and early interaction text embedding models?\",\n",
+    "]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "347ad924a3449743",
+   "metadata": {
+    "collapsed": false
+   },
+   "source": [
+    "*NOTE*: ColBERT computes query and documents embeddings differently, make sure to use the corresponding methods."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 17,
+   "id": "496fbf51e4eaaae",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-03T17:39:34.379885Z",
+     "start_time": "2024-06-03T17:39:34.316257Z"
+    },
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": [
+    "document_embeddings = list(\n",
+    "    embedding_model.embed(documents)\n",
+    ")  # embed and qury_embed return generators,\n",
+    "# which we need to evaluate by writing them to a list\n",
+    "query_embeddings = list(embedding_model.query_embed(queries))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 18,
+   "id": "50595bb0498f0c7c",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-03T17:39:34.793528Z",
+     "start_time": "2024-06-03T17:39:34.788545Z"
+    },
+    "collapsed": false
+   },
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "((26, 128), (32, 128))"
+      ]
+     },
+     "execution_count": 18,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "document_embeddings[0].shape, query_embeddings[0].shape"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "13e43f2c24a7d5fc",
+   "metadata": {
+    "collapsed": false
+   },
+   "source": [
+    "Don't worry about query embeddings having the bigger shape in this case. \n",
+    "ColBERT authors recommend to pad queries with [MASK] tokens to 32 tokens.\n",
+    "They also recommends to truncate queries to 32 tokens, however we don't do that in FastEmbed, so you can put some straight into the queries."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "bb1a4011effd3699",
+   "metadata": {
+    "collapsed": false
+   },
+   "source": [
+    "## MaxSim operator"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "e9ea4cf82521f2de",
+   "metadata": {
+    "collapsed": false
+   },
+   "source": [
+    "Qdrant will support ColBERT as of the next version (v1.10), however, at the moment, you can compute embedding similarities manually.   "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 19,
+   "id": "f84392f63d2c6076",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-03T17:39:36.431622Z",
+     "start_time": "2024-06-03T17:39:36.427363Z"
+    },
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": [
+    "import numpy as np\n",
+    "\n",
+    "\n",
+    "def compute_relevance_scores(\n",
+    "    query_embedding: np.array, document_embeddings: np.array, k: int\n",
+    ") -> list[int]:\n",
+    "    \"\"\"\n",
+    "    Compute relevance scores for top-k documents given a query.\n",
+    "\n",
+    "    :param query_embedding: Numpy array representing the query embedding, shape: [num_query_terms, embedding_dim]\n",
+    "    :param document_embeddings: Numpy array representing embeddings for documents, shape: [num_documents, max_doc_length, embedding_dim]\n",
+    "    :param k: Number of top documents to return\n",
+    "    :return: Indices of the top-k documents based on their relevance scores\n",
+    "    \"\"\"\n",
+    "    # Compute batch dot-product of query_embedding and document_embeddings\n",
+    "    # Resulting shape: [num_documents, num_query_terms, max_doc_length]\n",
+    "    scores = np.matmul(query_embedding, document_embeddings.transpose(0, 2, 1))\n",
+    "\n",
+    "    # Apply max-pooling across document terms (axis=2) to find the max similarity per query term\n",
+    "    # Shape after max-pool: [num_documents, num_query_terms]\n",
+    "    max_scores_per_query_term = np.max(scores, axis=2)\n",
+    "\n",
+    "    # Sum the scores across query terms to get the total score for each document\n",
+    "    # Shape after sum: [num_documents]\n",
+    "    total_scores = np.sum(max_scores_per_query_term, axis=1)\n",
+    "\n",
+    "    # Sort the documents based on their total scores and get the indices of the top-k documents\n",
+    "    sorted_indices = np.argsort(total_scores)[::-1][:k]\n",
+    "\n",
+    "    return sorted_indices"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 20,
+   "id": "c61d07bed7b60e35",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-03T17:39:37.053383Z",
+     "start_time": "2024-06-03T17:39:37.050926Z"
+    },
+    "collapsed": false
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Sorted document indices: [0 1]\n"
+     ]
+    }
+   ],
+   "source": [
+    "sorted_indices = compute_relevance_scores(\n",
+    "    np.array(query_embeddings[0]), np.array(document_embeddings), k=3\n",
+    ")\n",
+    "print(\"Sorted document indices:\", sorted_indices)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 22,
+   "id": "b24df2569970d9e8",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-03T17:40:52.276846Z",
+     "start_time": "2024-06-03T17:40:52.273789Z"
+    },
+    "collapsed": false
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Query: Are there any other late interaction text embedding models except ColBERT?\n",
+      "Document: ColBERT is a late interaction text embedding model, however, there are also other models such as TwinBERT.\n",
+      "Document: On the contrary to the late interaction models, the early interaction models contains interaction steps at embedding generation process\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(f\"Query: {queries[0]}\")\n",
+    "for index in sorted_indices:\n",
+    "    print(f\"Document: {documents[index]}\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "6de537c37aff3927",
+   "metadata": {
+    "collapsed": false
+   },
+   "source": [
+    "## Use-case recommendation"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "37e3525d3259cd2b",
+   "metadata": {
+    "collapsed": false
+   },
+   "source": [
+    "Despite ColBERT allows to compute embeddings independently and spare some workload offline, it still computes more resources than no interaction models. Due to this, it might be more reasonable to use ColBERT not as a first-stage retriever, but as a re-ranker.\n",
+    "\n",
+    "The first-stage retriever would then be a no-interaction model, which e.g. retrieves first 100 or 500 examples, and leave the final ranking to the ColBERT model."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "cfa922793454b4ad",
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 2
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython2",
+   "version": "2.7.6"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}
@@ -0,0 +1,540 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "ntGNDuSCeAR2"
+   },
+   "source": [
+    "# FastEmbed on GPU\n",
+    "\n",
+    "As of version 0.2.7 FastEmbed supports GPU acceleration.\n",
+    "\n",
+    "This notebook covers the installation process and usage of fastembed on GPU.\n",
+    "\n",
+    "## Installation\n",
+    "\n",
+    "Fastembed depends on `onnxruntime` and inherits its scheme of GPU support.\n",
+    "\n",
+    "In order to use GPU with onnx models, you would need to have `onnxruntime-gpu` package, which substitutes all the `onnxruntime` functionality.\n",
+    "Fastembed mimics this behavior and requires `fastembed-gpu` package to be installed."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "id": "GK2XADwUeEK7"
+   },
+   "outputs": [],
+   "source": [
+    "!pip install fastembed-gpu"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "3aiGPqjCeGzo"
+   },
+   "source": [
+    "**NOTE**: `onnxruntime-gpu` and `onnxruntime` can't be installed in the same environment. If you have `onnxruntime` installed, you would need to uninstall it before installing `onnxruntime-gpu`. Same is true for `fastembed` and `fastembed-gpu`."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "3xx3r-9jgAMi"
+   },
+   "source": [
+    "### CUDA 12.x support\n",
+    "You can check your CUDA version using such commands as `nvidia-smi` or `nvcc --version`\n",
+    "\n",
+    "Starting from version 1.19.0, onnxruntime-gpu ships with support for CUDA 12.x by default.\n",
+    "\n",
+    "Google Colab notebooks have by default CUDA 12.x and CuDNN 8.x.\n",
+    "\n",
+    "Latest version of `onnxruntime-gpu` requires CuDNN 9.x, in order to install it you can run the following command: "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!sudo apt install cudnn9\n",
+    "!pip install fastembed-gpu -qqq"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "If it necessary to work with CuDNN 8, you can consider locking `onnxruntime-gpu` to 1.18.0 with CUDA 12.x by this command:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!pip install onnxruntime-gpu==1.18.0 -i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/simple/ -qq\n",
+    "!pip install fastembed-gpu -qqq"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### CUDA 11.x support\n",
+    "To use latest version of `onnxruntime-gpu` with CUDA 11.x, you can run the following command:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!pip install onnxruntime-gpu -i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-11/pypi/simple/ -qq"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "**NOTE**: Ensure that CuDNN 9.x is installed when working with the latest `onnxruntime-gpu`, whether using CUDA 11.x or 12.x."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "id": "Igv5RXhSeO68"
+   },
+   "source": [
+    "### CUDA drivers\n",
+    "\n",
+    "FastEmbed does not include CUDA drivers and CuDNN libraries.\n",
+    "You would need to take care of the environment setup on your own.\n",
+    "The dependencies required for the chosen onnxruntime version are listed in the [CUDA Execution Provider requirements](https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements)."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Setting up fastembed-gpu on GCP\n",
+    "\n",
+    "#### CUDA drivers\n",
+    "[CUDA 11.8 toolkit](https://developer.nvidia.com/cuda-11-8-0-download-archive) or [CUDA 12.x toolkit](https://developer.nvidia.com/cuda-downloads) has to be installed if they haven't yet been set up.\n",
+    "\n",
+    "#### Example of setting up CUDA 12.x on Ubuntu 22.04\n",
+    "Make sure to download an archive which has been created for your particular platform, CPU architecture and OS distribution.\n",
+    "\n",
+    "For Ubuntu 22.04 with x86_64 CPU architecture the following [archive](https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=22.04&target_type=deb_network) has to be downloaded.\n",
+    "\n",
+    "```bash\n",
+    "wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb\n",
+    "sudo dpkg -i cuda-keyring_1.1-1_all.deb\n",
+    "sudo apt-get update\n",
+    "sudo apt-get -y install cuda\n",
+    "```\n",
+    "**NOTE**: Specific CUDA libraries can be found in the [meta packages section](https://docs.nvidia.com/cuda/cuda-installation-guide-linux/#meta-packages) in the CUDA installation guide.\n",
+    "\n",
+    "**NOTE**: When installing CUDA, the environment variable might not be set by default. Make sure to add the following line to your environment variables:\n",
+    "```bash\n",
+    "LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH\n",
+    "```\n",
+    "This will ensure that the CUDA libraries are properly linked.\n",
+    "\n",
+    "#### CuDNN 9.x\n",
+    "CuDNN 9.x library can be installed via the following [archive](https://developer.nvidia.com/rdp/cudnn-archive).\n",
+    "\n",
+    "#### Example of setting up CuDNN 9.x on Ubuntu 22.04\n",
+    "CuDNN 9.x for Ubuntu 22.04 x86_64 [archive](https://developer.nvidia.com/cudnn-downloads?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=22.04&target_type=deb_network) can be downloaded and installed in the following way:\n",
+    "```bash\n",
+    "wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb\n",
+    "sudo dpkg -i cuda-keyring_1.1-1_all.deb\n",
+    "sudo apt-get update\n",
+    "sudo apt-get -y install cudnn\n",
+    "```\n",
+    "**NOTE**: When installing CuDNN, you can choose specific version, cudnn-cuda-11 or cudnn-cuda-12"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Common issues\n",
+    "\n",
+    "The following are some common issues that may arise while using `fastembed-gpu` if not installed properly:\n",
+    "\n",
+    "CUDA library is not installed:\n",
+    "```bash\n",
+    "FAIL : Failed to load library libonnxruntime_providers_cuda.so with error: libcublasLt.so.x: cannot open shared object file: No such file or directory\n",
+    "```\n",
+    "\n",
+    "\n",
+    "CuDNN library is not installed:\n",
+    "```bash\n",
+    "FAIL : Failed to load library libonnxruntime_providers_cuda.so with error: libcudnn.so.x: cannot open shared object file: No such file or directory\n",
+    "```\n",
+    "\n",
+    "\n",
+    "CUDA library path is not set:\n",
+    "```bash\n",
+    "FAIL : Failed to load library libonnxruntime_providers_cuda.so with error: libcufft.so.x: failed to map segment from shared object\n",
+    "```\n",
+    "\n",
+    "Make sure to add the following line to your environment variables:\n",
+    "```bash\n",
+    "LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH\n",
+    "```"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Usage"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {
+    "colab": {
+     "base_uri": "https://localhost:8080/",
+     "height": 334,
+     "referenced_widgets": [
+      "aacf08a7aa444b64a2efad1967d28a53",
+      "5606aa785de74d65a9928b31c0be8a53",
+      "d4ec9d3b74ec4412894da2161ed2bddf",
+      "8edd544c3e074ec1813e5b9d1aef43d9",
+      "9898890f8a75468ea20e3ce319d0b6e2",
+      "da3b18abb16241a0a7191ee9afcb0510",
+      "258a619168824253a6a329efdc51ebe6",
+      "53c7cdc967d24faba0b5c659c94c50b8",
+      "e9348d8be28d408e8e760c71b21ab294",
+      "0ba06e0816714f2fbdec8260f160abc0",
+      "0b96563334964d449dd34f35b6b3e715",
+      "11c2eec490e8479b944eec7f30cb1ca2",
+      "91463da0d1c5466795e06ab586002259",
+      "30f4f7833406474f89ef0700b00a33aa",
+      "4302c304ec6a4b5985797e300bd7e353",
+      "2605640c7b824ed7aa137d404e14b774",
+      "b02efe3a33d04f06aa8938719ab35671",
+      "50408e5d052343b1a1b44a0fae0f801d",
+      "1be01c95d9e84f8ea88367c987a72fdc",
+      "a109c13bc93a449186424542dc330be8",
+      "4adce304ce1947b5a01dde10bbb3bb8c",
+      "a761366a37e44837a25e0f25b18efed2",
+      "94512b9055e546389471197b76ad5449",
+      "072dca00bd7b4918a178f90ccabf698a",
+      "48a856c59ef74cc3834521b1bf616541",
+      "c020c503aeaa464cad643ade5ee3ae24",
+      "a4e7e40c0bbd4f878c20a9f65fe3a048",
+      "e8c0a1c339fd47668d944a9defad79d4",
+      "cd782d35c6bd40c0a60d57b1828a7251",
+      "04f638ab08da4d20928644c4ba03f8ef",
+      "17f20477fc79475f97adf1c1f64a4192",
+      "96f7b5a2e224462e9fcffd03f906a593",
+      "755cd32d9fc9407c80a160f45c802d1e",
+      "a886258e7cd14c048b58391d7b772901",
+      "bc3e48f826a74840867a6209e622b75e",
+      "125b2ac0f78043bba7eca53474ca44c4",
+      "82f186d1ffb4435d94a6c7e9025242ef",
+      "77000333e5ca4094be291ad82d4a627a",
+      "7fe64fb53055431488d002c76c8e331e",
+      "7de59ae9919f4a5bb2b6e601a3c02412",
+      "97a69423a6644eab87fc636e182f23a4",
+      "4df936d1065b41f4bf02ed394fdf7b7e",
+      "3918bd1affa3454e8e9044a418a056ea",
+      "163b27ae0bce41e5b48efcb4b3fd780d",
+      "94631fd6e0744085bc79c3121de4a9f7",
+      "31cd98d66bc54418b35e70fbbc0fa3c0",
+      "6d21627a638b4ddca6fe7bfb80a621b5",
+      "b37bed9dc4fe45c08b8397288fe5b1a9",
+      "164fef95d1414177a40d563f5682f6a3",
+      "1a9a0ea53448413a8e4b360b7bb69e26",
+      "dd1a4483b4b045c6929e3d2cf1338f63",
+      "496ddd8e05f949cd8cbba8e677f476ac",
+      "2813be951d7f48b2aad1dd4a444ce3eb",
+      "8e9a2c2dd21942edbdfecb3b7dffc70b",
+      "08a10fe247f1425db044cfc13f2fb384",
+      "b8786aded92d421592bc7623c5c7899e",
+      "c91a20a9433d4016ba2db69fa50e0b4d",
+      "e997820738594c6dadb061908d7afdc1",
+      "a5fc751f81ae498f9aa55ece0e6853b2",
+      "2aee4fc8cda64c5eb8722be81e48e0ca",
+      "3a53e8624dff48b3959875ef58ee99ce",
+      "50a70044f77542108fe188598e70797e",
+      "13cf998b35ae4507a63e797f6fa3eada",
+      "6209eb6a68cf4a378767ef34d0d9216d",
+      "7395db766b944af9b41d6b56c9ada0b1",
+      "42122c317ec648688f0164a1adb5df28"
+     ]
+    },
+    "id": "Ttf4YggPeQQK",
+    "outputId": "aa75129d-9e2d-4c88-cf03-251dd43a11b1"
+   },
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/usr/local/lib/python3.10/dist-packages/huggingface_hub/utils/_token.py:88: UserWarning: \n",
+      "The secret `HF_TOKEN` does not exist in your Colab secrets.\n",
+      "To authenticate with the Hugging Face Hub, create a token in your settings tab (https://huggingface.co/settings/tokens), set it as secret in your Google Colab and restart your session.\n",
+      "You will be able to reuse this secret in all of your notebooks.\n",
+      "Please note that authentication is recommended but still optional to access public models or datasets.\n",
+      "  warnings.warn(\n"
+     ]
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "aacf08a7aa444b64a2efad1967d28a53",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Fetching 5 files:   0%|          | 0/5 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "11c2eec490e8479b944eec7f30cb1ca2",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "tokenizer_config.json:   0%|          | 0.00/1.24k [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "94512b9055e546389471197b76ad5449",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "config.json:   0%|          | 0.00/706 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "a886258e7cd14c048b58391d7b772901",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "special_tokens_map.json:   0%|          | 0.00/695 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "94631fd6e0744085bc79c3121de4a9f7",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "tokenizer.json:   0%|          | 0.00/711k [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "b8786aded92d421592bc7623c5c7899e",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "model_optimized.onnx:   0%|          | 0.00/66.5M [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/plain": [
+       "['CUDAExecutionProvider', 'CPUExecutionProvider']"
+      ]
+     },
+     "execution_count": 2,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "import numpy as np\n",
+    "\n",
+    "from fastembed import TextEmbedding\n",
+    "\n",
+    "embedding_model_gpu = TextEmbedding(\n",
+    "    model_name=\"BAAI/bge-small-en-v1.5\", providers=[\"CUDAExecutionProvider\"]\n",
+    ")\n",
+    "embedding_model_gpu.model.model.get_providers()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {
+    "id": "iPtoHf7GeV-i"
+   },
+   "outputs": [],
+   "source": "documents: list[str] = list(np.repeat(\"Demonstrating GPU acceleration in fastembed\", 500))"
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {
+    "colab": {
+     "base_uri": "https://localhost:8080/"
+    },
+    "id": "islhyLf4ed-H",
+    "outputId": "8c8ed09b-9eac-438f-97bc-578751975148"
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "43.4 ms ± 2.06 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
+     ]
+    }
+   ],
+   "source": [
+    "%%timeit\n",
+    "list(embedding_model_gpu.embed(documents))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {
+    "colab": {
+     "base_uri": "https://localhost:8080/",
+     "height": 67,
+     "referenced_widgets": [
+      "9c306ce5188c45feb8dfb9089592591c",
+      "296ff54c6e61441f978084df59626598",
+      "d6d42b4f245a49b7ba7769e23a3202fc",
+      "39ce7754480147759c16a3089d8105af",
+      "8253960a069d4106863a75faae54b90d",
+      "7ccf959452af4c0b873c7567747f0816",
+      "ac9d0b5a5b1f401e90a1cc9ffe6d4b4c",
+      "0aada067dec3472f9aba1772d6b775a5",
+      "07597b1287e04653b80c47a771549376",
+      "054be1dd9f084cae911745b692ccd929",
+      "ab19e8e831694e308a4b79f05aff728e"
+     ]
+    },
+    "id": "bOKVUvWJegYJ",
+    "outputId": "dde74917-08b0-4ce2-9a2b-cc31e02cafb2"
+   },
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "9c306ce5188c45feb8dfb9089592591c",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Fetching 5 files:   0%|          | 0/5 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/plain": [
+       "['CPUExecutionProvider']"
+      ]
+     },
+     "execution_count": 5,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "embedding_model_cpu = TextEmbedding(model_name=\"BAAI/bge-small-en-v1.5\")\n",
+    "embedding_model_cpu.model.model.get_providers()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {
+    "colab": {
+     "base_uri": "https://localhost:8080/"
+    },
+    "id": "0NJj9RvSfASP",
+    "outputId": "526f5280-99bd-454e-8af8-6a860ad96e54"
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "4.33 s ± 591 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
+     ]
+    }
+   ],
+   "source": [
+    "%%timeit\n",
+    "list(embedding_model_cpu.embed(documents))"
+   ]
+  }
+ ],
+ "metadata": {
+  "accelerator": "GPU",
+  "colab": {
+   "gpuType": "T4",
+   "provenance": []
+  },
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.12"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 1
+}
@@ -0,0 +1,88 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Fastembed Multi-GPU Tutorial\n",
+    "This tutorial demonstrates how to leverage multi-GPU support in Fastembed. Fastembed supports embedding text and images utilizing modern GPUs for acceleration. Let's explore how to use Fastembed with multiple GPUs step by step."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "#### Prerequisites\n",
+    "To get started, ensure you have the following installed:\n",
+    "- Python 3.9 or later\n",
+    "- Fastembed (`pip install fastembed-gpu`)\n",
+    "- Refer to [this](https://github.com/qdrant/fastembed/blob/main/docs/examples/FastEmbed_GPU.ipynb) tutorial if you have issues with GPU dependencies\n",
+    "- Access to a multi-GPU server"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Multi-GPU using cuda argument with TextEmbedding Model"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from fastembed import TextEmbedding\n",
+    "\n",
+    "# define the documents to embed\n",
+    "docs = [\"hello world\", \"flag embedding\"] * 100\n",
+    "\n",
+    "# define gpu ids\n",
+    "device_ids = [0, 1]\n",
+    "\n",
+    "if __name__ == \"__main__\":\n",
+    "    # initialize a TextEmbedding model using CUDA\n",
+    "    text_model = TextEmbedding(\n",
+    "        model_name=\"sentence-transformers/all-MiniLM-L6-v2\",\n",
+    "        cuda=True,\n",
+    "        device_ids=device_ids,\n",
+    "        lazy_load=True,\n",
+    "    )\n",
+    "\n",
+    "    # generate embeddings\n",
+    "    text_embeddings = list(text_model.embed(docs, batch_size=2, parallel=len(device_ids)))\n",
+    "    print(text_embeddings)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "In this snippet:\n",
+    "- `cuda=True` enables GPU acceleration.\n",
+    "- `device_ids=[0, 1]` specifies GPUs to use. Replace `[0, 1]` with available GPU IDs.\n",
+    "- `lazy_load=True`\n",
+    "\n",
+    "**NOTE**: When using multi-GPU settings, it is important to configure `parallel` and `lazy_load` properly to avoid inefficiencies:\n",
+    "\n",
+    "`parallel`: This parameter enables multi-GPU support by spawning child processes for each GPU specified in device_ids. To ensure proper utilization, the value of `parallel` must match the number of GPUs in device_ids. If using a single GPU, this parameter is not necessary.\n",
+    "\n",
+    "`lazy_load`: Enabling `lazy_load` prevents redundant memory usage. Without `lazy_load`, the model is initially loaded into the memory of the first GPU by the main process. When child processes are spawned for each GPU, the model is reloaded on the first GPU, causing redundant memory consumption and inefficiencies."
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": ".venv",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "name": "python",
+   "version": "3.10.15"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}
@@ -0,0 +1,128 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "aa0a86859809102",
+   "metadata": {
+    "collapsed": false
+   },
+   "source": [
+    "# Image Embedding\n",
+    "As of version 0.3.0 fastembed supports computation of image embeddings.\n",
+    "\n",
+    "The process is as easy and straightforward as with text embeddings. Let's see how it works."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "id": "cea8fd5c019571fe",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-02T11:35:40.126023Z",
+     "start_time": "2024-06-02T11:35:39.864701Z"
+    },
+    "collapsed": false
+   },
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Fetching 3 files: 100%|██████████| 3/3 [00:00<00:00, 47482.69it/s]\n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": "[array([0.        , 0.        , 0.        , ..., 0.        , 0.01139933,\n        0.        ], dtype=float32),\n array([0.02169187, 0.        , 0.        , ..., 0.        , 0.00848291,\n        0.        ], dtype=float32)]"
+     },
+     "execution_count": 5,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "from fastembed import ImageEmbedding\n",
+    "\n",
+    "model = ImageEmbedding(\"Qdrant/resnet50-onnx\")\n",
+    "\n",
+    "embeddings_generator = model.embed(\n",
+    "    [\"../../tests/misc/image.jpeg\", \"../../tests/misc/small_image.jpeg\"]\n",
+    ")\n",
+    "embeddings_list = list(embeddings_generator)\n",
+    "embeddings_list"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "3f838f18523ad1e0",
+   "metadata": {
+    "collapsed": false
+   },
+   "source": [
+    "## Preprocessing\n",
+    "\n",
+    "Preprocessing is encapsulated in the ImageEmbedding class, applied operations are identical to the ones provided by [Hugging Face Transformers](https://huggingface.co/docs/transformers/en/index).\n",
+    "You don't need to think about batching, opening/closing files, resizing images, etc., Fastembed will take care of it."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "894b33ff9b385d72",
+   "metadata": {
+    "collapsed": false
+   },
+   "source": [
+    "## Supported models\n",
+    "\n",
+    "List of supported image embedding models can either be found [here](https://qdrant.github.io/fastembed/examples/Supported_Models/#supported-image-embedding-models) or by calling the `ImageEmbedding.list_supported_models()` method."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "id": "6d6a4cbbd2200d14",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-02T11:40:19.313226Z",
+     "start_time": "2024-06-02T11:40:19.309845Z"
+    },
+    "collapsed": false
+   },
+   "outputs": [
+    {
+     "data": {
+      "text/plain": "[{'model': 'Qdrant/clip-ViT-B-32-vision',\n  'dim': 512,\n  'description': 'CLIP vision encoder based on ViT-B/32',\n  'size_in_GB': 0.34,\n  'sources': {'hf': 'Qdrant/clip-ViT-B-32-vision'},\n  'model_file': 'model.onnx'},\n {'model': 'Qdrant/resnet50-onnx',\n  'dim': 2048,\n  'description': 'ResNet-50 from `Deep Residual Learning for Image Recognition <https://arxiv.org/abs/1512.03385>`__.',\n  'size_in_GB': 0.1,\n  'sources': {'hf': 'Qdrant/resnet50-onnx'},\n  'model_file': 'model.onnx'}]"
+     },
+     "execution_count": 6,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "ImageEmbedding.list_supported_models()"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 2
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython2",
+   "version": "2.7.6"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}
@@ -0,0 +1,389 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Introduction to SPLADE with FastEmbed\n",
+    "\n",
+    "In this notebook, we will explore how to generate Sparse Vectors -- in particular a variant of the [SPLADE](https://arxiv.org/abs/2107.05720).\n",
+    "\n",
+    "> 💡 The original [naver/SPLADE](https://github.com/naver/splade) models were licensed CC BY-NC-SA 4.0 -- Not for Commercial Use. This [SPLADE++](https://huggingface.co/prithivida/Splade_PP_en_v1) model is Apache License and hence, licensed for commercial use. \n",
+    "\n",
+    "## Outline:\n",
+    "1. [What is SPLADE?](#What-is-SPLADE?)\n",
+    "2. [Setting up the environment](#Setting-up-the-environment)\n",
+    "3. [Generating SPLADE vectors with FastEmbed](#Generating-SPLADE-vectors-with-FastEmbed)\n",
+    "4. [Understanding SPLADE vectors](#Understanding-SPLADE-vectors)\n",
+    "5. [Observations and Design Choices](#Observations-and-Model-Design-Choices)\n",
+    "\n",
+    "\n",
+    "## What is SPLADE?\n",
+    "\n",
+    "SPLADE was a novel method for _learning_ sparse vectors for text representation. This model beats BM25 -- the underlying approach for the Elastic/Lucene family of implementations. Thus making it highly effective for tasks such as information retrieval, document classification, and more. \n",
+    "\n",
+    "The key advantage of SPLADE is its ability to generate sparse vectors, which are more efficient and interpretable than dense vectors. This makes SPLADE a powerful tool for handling large-scale text data.\n",
+    "\n",
+    "## Setting up the environment\n",
+    "\n",
+    "This notebook uses few dependencies, which are installed below: "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# !pip install -q fastembed"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Let's get started! 🚀"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-03-30T00:49:20.516644Z",
+     "start_time": "2024-03-30T00:49:20.188543Z"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "from fastembed import SparseTextEmbedding, SparseEmbedding"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "> You can find the list of all supported Sparse Embedding models by calling this API: `SparseTextEmbedding.list_supported_models()`"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-03-30T00:49:22.366294Z",
+     "start_time": "2024-03-30T00:49:22.362384Z"
+    }
+   },
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "[{'model': 'prithvida/Splade_PP_en_v1',\n",
+       "  'vocab_size': 30522,\n",
+       "  'description': 'Misspelled version of the model. Retained for backward compatibility. Independent Implementation of SPLADE++ Model for English',\n",
+       "  'size_in_GB': 0.532,\n",
+       "  'sources': {'hf': 'Qdrant/SPLADE_PP_en_v1'}},\n",
+       " {'model': 'prithivida/Splade_PP_en_v1',\n",
+       "  'vocab_size': 30522,\n",
+       "  'description': 'Independent Implementation of SPLADE++ Model for English',\n",
+       "  'size_in_GB': 0.532,\n",
+       "  'sources': {'hf': 'Qdrant/SPLADE_PP_en_v1'}}]"
+      ]
+     },
+     "execution_count": 3,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "SparseTextEmbedding.list_supported_models()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-03-30T00:49:27.193530Z",
+     "start_time": "2024-03-30T00:49:26.139248Z"
+    }
+   },
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "2aa47b26ab01475e8d3577433037f685",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Fetching 9 files:   0%|          | 0/9 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "model_name = \"prithvida/Splade_PP_en_v1\"\n",
+    "# This triggers the model download\n",
+    "model = SparseTextEmbedding(model_name=model_name)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-03-30T00:49:28.624109Z",
+     "start_time": "2024-03-30T00:49:28.399960Z"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "documents: list[str] = [\n",
+    "    \"Chandrayaan-3 is India's third lunar mission\",\n",
+    "    \"It aimed to land a rover on the Moon's surface - joining the US, China and Russia\",\n",
+    "    \"The mission is a follow-up to Chandrayaan-2, which had partial success\",\n",
+    "    \"Chandrayaan-3 will be launched by the Indian Space Research Organisation (ISRO)\",\n",
+    "    \"The estimated cost of the mission is around $35 million\",\n",
+    "    \"It will carry instruments to study the lunar surface and atmosphere\",\n",
+    "    \"Chandrayaan-3 landed on the Moon's surface on 23rd August 2023\",\n",
+    "    \"It consists of a lander named Vikram and a rover named Pragyan similar to Chandrayaan-2. Its propulsion module would act like an orbiter.\",\n",
+    "    \"The propulsion module carries the lander and rover configuration until the spacecraft is in a 100-kilometre (62 mi) lunar orbit\",\n",
+    "    \"The mission used GSLV Mk III rocket for its launch\",\n",
+    "    \"Chandrayaan-3 was launched from the Satish Dhawan Space Centre in Sriharikota\",\n",
+    "    \"Chandrayaan-3 was launched earlier in the year 2023\",\n",
+    "]\n",
+    "sparse_embeddings_list: list[SparseEmbedding] = list(\n",
+    "    model.embed(documents, batch_size=6)\n",
+    ")  # batch_size is optional, notice the generator"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-03-30T00:49:29.646340Z",
+     "start_time": "2024-03-30T00:49:29.643411Z"
+    }
+   },
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "SparseEmbedding(values=array([0.05297208, 0.01963477, 0.36459631, 1.38508618, 0.71776593,\n",
+       "       0.12667948, 0.46230844, 0.446771  , 0.26897505, 1.01519883,\n",
+       "       1.5655334 , 0.29412213, 1.53102326, 0.59785569, 1.1001817 ,\n",
+       "       0.02079751, 0.09955651, 0.44249091, 0.09747757, 1.53519952,\n",
+       "       1.36765671, 0.15740395, 0.49882549, 0.38629025, 0.76612782,\n",
+       "       1.25805044, 0.39058095, 0.27236196, 0.45152301, 0.48262018,\n",
+       "       0.26085234, 1.35912788, 0.70710695, 1.71639752]), indices=array([ 1010,  1011,  1016,  1017,  2001,  2018,  2034,  2093,  2117,\n",
+       "        2319,  2353,  2509,  2634,  2686,  2796,  2817,  2922,  2959,\n",
+       "        3003,  3148,  3260,  3390,  3462,  3523,  3822,  4231,  4316,\n",
+       "        4774,  5590,  5871,  6416, 11926, 12076, 16469]))"
+      ]
+     },
+     "execution_count": 6,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "index = 0\n",
+    "sparse_embeddings_list[index]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "The previous output is a SparseEmbedding object for the first document in our list.\n",
+    "\n",
+    "It contains two arrays: values and indices. \n",
+    "- The 'values' array represents the weights of the features (tokens) in the document.\n",
+    "- The 'indices' array represents the indices of these features in the model's vocabulary.\n",
+    "\n",
+    "Each pair of corresponding values and indices represents a token and its weight in the document."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-03-30T00:49:31.549533Z",
+     "start_time": "2024-03-30T00:49:31.546398Z"
+    }
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Token at index 1010 has weight 0.05297207832336426\n",
+      "Token at index 1011 has weight 0.01963476650416851\n",
+      "Token at index 1016 has weight 0.36459630727767944\n",
+      "Token at index 1017 has weight 1.385086178779602\n",
+      "Token at index 2001 has weight 0.7177659273147583\n"
+     ]
+    }
+   ],
+   "source": [
+    "# Let's print the first 5 features and their weights for better understanding.\n",
+    "for i in range(5):\n",
+    "    print(\n",
+    "        f\"Token at index {sparse_embeddings_list[0].indices[i]} has weight {sparse_embeddings_list[0].values[i]}\"\n",
+    "    )"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Understanding SPLADE vectors\n",
+    "\n",
+    "This is still a little abstract, so let's use the tokenizer vocab to make sense of these indices."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-03-30T00:49:36.203640Z",
+     "start_time": "2024-03-30T00:49:34.889654Z"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "import json\n",
+    "from transformers import AutoTokenizer\n",
+    "\n",
+    "tokenizer = AutoTokenizer.from_pretrained(\n",
+    "    SparseTextEmbedding.list_supported_models()[0][\"sources\"][\"hf\"]\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-03-30T00:49:36.210049Z",
+     "start_time": "2024-03-30T00:49:36.206825Z"
+    }
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "{\n",
+      "    \"chandra\": 1.7163975238800049,\n",
+      "    \"third\": 1.5655333995819092,\n",
+      "    \"##ya\": 1.535199522972107,\n",
+      "    \"india\": 1.5310232639312744,\n",
+      "    \"3\": 1.385086178779602,\n",
+      "    \"mission\": 1.3676567077636719,\n",
+      "    \"lunar\": 1.3591278791427612,\n",
+      "    \"moon\": 1.2580504417419434,\n",
+      "    \"indian\": 1.1001816987991333,\n",
+      "    \"##an\": 1.015198826789856,\n",
+      "    \"3rd\": 0.7661278247833252,\n",
+      "    \"was\": 0.7177659273147583,\n",
+      "    \"spacecraft\": 0.7071069478988647,\n",
+      "    \"space\": 0.5978556871414185,\n",
+      "    \"flight\": 0.4988254904747009,\n",
+      "    \"satellite\": 0.4826201796531677,\n",
+      "    \"first\": 0.46230843663215637,\n",
+      "    \"expedition\": 0.4515230059623718,\n",
+      "    \"three\": 0.4467709958553314,\n",
+      "    \"fourth\": 0.44249090552330017,\n",
+      "    \"vehicle\": 0.390580952167511,\n",
+      "    \"iii\": 0.3862902522087097,\n",
+      "    \"2\": 0.36459630727767944,\n",
+      "    \"##3\": 0.2941221296787262,\n",
+      "    \"planet\": 0.27236196398735046,\n",
+      "    \"second\": 0.26897504925727844,\n",
+      "    \"missions\": 0.2608523368835449,\n",
+      "    \"launched\": 0.15740394592285156,\n",
+      "    \"had\": 0.12667948007583618,\n",
+      "    \"largest\": 0.09955651313066483,\n",
+      "    \"leader\": 0.09747757017612457,\n",
+      "    \",\": 0.05297207832336426,\n",
+      "    \"study\": 0.02079751156270504,\n",
+      "    \"-\": 0.01963476650416851\n",
+      "}\n"
+     ]
+    }
+   ],
+   "source": [
+    "def get_tokens_and_weights(sparse_embedding, tokenizer):\n",
+    "    token_weight_dict = {}\n",
+    "    for i in range(len(sparse_embedding.indices)):\n",
+    "        token = tokenizer.decode([sparse_embedding.indices[i]])\n",
+    "        weight = sparse_embedding.values[i]\n",
+    "        token_weight_dict[token] = weight\n",
+    "\n",
+    "    # Sort the dictionary by weights\n",
+    "    token_weight_dict = dict(\n",
+    "        sorted(token_weight_dict.items(), key=lambda item: item[1], reverse=True)\n",
+    "    )\n",
+    "    return token_weight_dict\n",
+    "\n",
+    "\n",
+    "# Test the function with the first SparseEmbedding\n",
+    "print(json.dumps(get_tokens_and_weights(sparse_embeddings_list[index], tokenizer), indent=4))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Observations and Model Design Choices\n",
+    "\n",
+    "1. The relative order of importance is quite useful. The most important tokens in the sentence have the highest weights.\n",
+    "1. **Term Expansion**: The model can expand the terms in the document. This means that the model can generate weights for tokens that are not present in the document but are related to the tokens in the document. This is a powerful feature that allows the model to capture the context of the document. Here, you'll see that the model has added the tokens '3' from 'third' and 'moon' from 'lunar' to the sparse vector.\n",
+    "\n",
+    "### Design Choices\n",
+    "\n",
+    "1. The weights are not normalized. This means that the sum of the weights is not 1 or 100. This is a common practice in sparse embeddings, as it allows the model to capture the importance of each token in the document.\n",
+    "1. Tokens are included in the sparse vector only if they are present in the model's vocabulary. This means that the model will not generate a weight for tokens that it has not seen during training.\n",
+    "1. Tokens do not map to words directly -- allowing you to gracefully handle typo errors and out-of-vocabulary tokens."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "fst",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.13"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}
@@ -0,0 +1,874 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-11-13T09:01:03.324551Z",
+     "start_time": "2024-11-13T09:01:03.234711Z"
+    }
+   },
+   "source": [
+    "%load_ext autoreload\n",
+    "%autoreload 2"
+   ],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "The autoreload extension is already loaded. To reload it, use:\n",
+      "  %reload_ext autoreload\n"
+     ]
+    }
+   ],
+   "execution_count": 10
+  },
+  {
+   "cell_type": "code",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-11-13T09:01:04.505772Z",
+     "start_time": "2024-11-13T09:01:04.493296Z"
+    }
+   },
+   "source": [
+    "import pandas as pd\n",
+    "\n",
+    "from fastembed import (\n",
+    "    SparseTextEmbedding,\n",
+    "    TextEmbedding,\n",
+    "    LateInteractionTextEmbedding,\n",
+    "    ImageEmbedding,\n",
+    ")\n",
+    "from fastembed.rerank.cross_encoder import TextCrossEncoder"
+   ],
+   "outputs": [],
+   "execution_count": 11
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Supported Text Embedding Models"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-11-13T09:01:05.812271Z",
+     "start_time": "2024-11-13T09:01:05.795846Z"
+    }
+   },
+   "source": [
+    "supported_models = (\n",
+    "    pd.DataFrame(TextEmbedding.list_supported_models())\n",
+    "    .sort_values(\"size_in_GB\")\n",
+    "    .drop(columns=[\"sources\", \"model_file\", \"additional_files\"])\n",
+    "    .reset_index(drop=True)\n",
+    ")\n",
+    "supported_models"
+   ],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "                                                model   dim  \\\n",
+       "0                              BAAI/bge-small-en-v1.5   384   \n",
+       "1                              BAAI/bge-small-zh-v1.5   512   \n",
+       "2                 snowflake/snowflake-arctic-embed-xs   384   \n",
+       "3              sentence-transformers/all-MiniLM-L6-v2   384   \n",
+       "4                  jinaai/jina-embeddings-v2-small-en   512   \n",
+       "5                                   BAAI/bge-small-en   384   \n",
+       "6                  snowflake/snowflake-arctic-embed-s   384   \n",
+       "7                    nomic-ai/nomic-embed-text-v1.5-Q   768   \n",
+       "8                               BAAI/bge-base-en-v1.5   768   \n",
+       "9   sentence-transformers/paraphrase-multilingual-...   384   \n",
+       "10                          Qdrant/clip-ViT-B-32-text   512   \n",
+       "11                  jinaai/jina-embeddings-v2-base-de   768   \n",
+       "12                                   BAAI/bge-base-en   768   \n",
+       "13                 snowflake/snowflake-arctic-embed-m   768   \n",
+       "14                     nomic-ai/nomic-embed-text-v1.5   768   \n",
+       "15                  jinaai/jina-embeddings-v2-base-en   768   \n",
+       "16                       nomic-ai/nomic-embed-text-v1   768   \n",
+       "17            snowflake/snowflake-arctic-embed-m-long   768   \n",
+       "18                 mixedbread-ai/mxbai-embed-large-v1  1024   \n",
+       "19                jinaai/jina-embeddings-v2-base-code   768   \n",
+       "20  sentence-transformers/paraphrase-multilingual-...   768   \n",
+       "21                 snowflake/snowflake-arctic-embed-l  1024   \n",
+       "22                                 thenlper/gte-large  1024   \n",
+       "23                             BAAI/bge-large-en-v1.5  1024   \n",
+       "24                     intfloat/multilingual-e5-large  1024   \n",
+       "\n",
+       "                                          description     license  size_in_GB  \n",
+       "0   Text embeddings, Unimodal (text), English, 512...         mit       0.067  \n",
+       "1   Text embeddings, Unimodal (text), Chinese, 512...         mit       0.090  \n",
+       "2   Text embeddings, Unimodal (text), English, 512...  apache-2.0       0.090  \n",
+       "3   Text embeddings, Unimodal (text), English, 256...  apache-2.0       0.090  \n",
+       "4   Text embeddings, Unimodal (text), English, 819...  apache-2.0       0.120  \n",
+       "5   Text embeddings, Unimodal (text), English, 512...         mit       0.130  \n",
+       "6   Text embeddings, Unimodal (text), English, 512...  apache-2.0       0.130  \n",
+       "7   Text embeddings, Multimodal (text, image), Eng...  apache-2.0       0.130  \n",
+       "8   Text embeddings, Unimodal (text), English, 512...         mit       0.210  \n",
+       "9   Text embeddings, Unimodal (text), Multilingual...  apache-2.0       0.220  \n",
+       "10  Text embeddings, Multimodal (text&image), Engl...         mit       0.250  \n",
+       "11  Text embeddings, Unimodal (text), Multilingual...  apache-2.0       0.320  \n",
+       "12  Text embeddings, Unimodal (text), English, 512...         mit       0.420  \n",
+       "13  Text embeddings, Unimodal (text), English, 512...  apache-2.0       0.430  \n",
+       "14  Text embeddings, Multimodal (text, image), Eng...  apache-2.0       0.520  \n",
+       "15  Text embeddings, Unimodal (text), English, 819...  apache-2.0       0.520  \n",
+       "16  Text embeddings, Multimodal (text, image), Eng...  apache-2.0       0.520  \n",
+       "17  Text embeddings, Unimodal (text), English, 204...  apache-2.0       0.540  \n",
+       "18  Text embeddings, Unimodal (text), English, 512...  apache-2.0       0.640  \n",
+       "19  Text embeddings, Unimodal (text), Multilingual...  apache-2.0       0.640  \n",
+       "20  Text embeddings, Unimodal (text), Multilingual...  apache-2.0       1.000  \n",
+       "21  Text embeddings, Unimodal (text), English, 512...  apache-2.0       1.020  \n",
+       "22  Text embeddings, Unimodal (text), English, 512...         mit       1.200  \n",
+       "23  Text embeddings, Unimodal (text), English, 512...         mit       1.200  \n",
+       "24  Text embeddings, Unimodal (text), Multilingual...         mit       2.240  "
+      ],
+      "text/html": [
+       "<div>\n",
+       "<style scoped>\n",
+       "    .dataframe tbody tr th:only-of-type {\n",
+       "        vertical-align: middle;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe tbody tr th {\n",
+       "        vertical-align: top;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe thead th {\n",
+       "        text-align: right;\n",
+       "    }\n",
+       "</style>\n",
+       "<table border=\"1\" class=\"dataframe\">\n",
+       "  <thead>\n",
+       "    <tr style=\"text-align: right;\">\n",
+       "      <th></th>\n",
+       "      <th>model</th>\n",
+       "      <th>dim</th>\n",
+       "      <th>description</th>\n",
+       "      <th>license</th>\n",
+       "      <th>size_in_GB</th>\n",
+       "    </tr>\n",
+       "  </thead>\n",
+       "  <tbody>\n",
+       "    <tr>\n",
+       "      <th>0</th>\n",
+       "      <td>BAAI/bge-small-en-v1.5</td>\n",
+       "      <td>384</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 512...</td>\n",
+       "      <td>mit</td>\n",
+       "      <td>0.067</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>1</th>\n",
+       "      <td>BAAI/bge-small-zh-v1.5</td>\n",
+       "      <td>512</td>\n",
+       "      <td>Text embeddings, Unimodal (text), Chinese, 512...</td>\n",
+       "      <td>mit</td>\n",
+       "      <td>0.090</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>2</th>\n",
+       "      <td>snowflake/snowflake-arctic-embed-xs</td>\n",
+       "      <td>384</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 512...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.090</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>3</th>\n",
+       "      <td>sentence-transformers/all-MiniLM-L6-v2</td>\n",
+       "      <td>384</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 256...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.090</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>4</th>\n",
+       "      <td>jinaai/jina-embeddings-v2-small-en</td>\n",
+       "      <td>512</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 819...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.120</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>5</th>\n",
+       "      <td>BAAI/bge-small-en</td>\n",
+       "      <td>384</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 512...</td>\n",
+       "      <td>mit</td>\n",
+       "      <td>0.130</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>6</th>\n",
+       "      <td>snowflake/snowflake-arctic-embed-s</td>\n",
+       "      <td>384</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 512...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.130</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>7</th>\n",
+       "      <td>nomic-ai/nomic-embed-text-v1.5-Q</td>\n",
+       "      <td>768</td>\n",
+       "      <td>Text embeddings, Multimodal (text, image), Eng...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.130</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>8</th>\n",
+       "      <td>BAAI/bge-base-en-v1.5</td>\n",
+       "      <td>768</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 512...</td>\n",
+       "      <td>mit</td>\n",
+       "      <td>0.210</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>9</th>\n",
+       "      <td>sentence-transformers/paraphrase-multilingual-...</td>\n",
+       "      <td>384</td>\n",
+       "      <td>Text embeddings, Unimodal (text), Multilingual...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.220</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>10</th>\n",
+       "      <td>Qdrant/clip-ViT-B-32-text</td>\n",
+       "      <td>512</td>\n",
+       "      <td>Text embeddings, Multimodal (text&amp;image), Engl...</td>\n",
+       "      <td>mit</td>\n",
+       "      <td>0.250</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>11</th>\n",
+       "      <td>jinaai/jina-embeddings-v2-base-de</td>\n",
+       "      <td>768</td>\n",
+       "      <td>Text embeddings, Unimodal (text), Multilingual...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.320</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>12</th>\n",
+       "      <td>BAAI/bge-base-en</td>\n",
+       "      <td>768</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 512...</td>\n",
+       "      <td>mit</td>\n",
+       "      <td>0.420</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>13</th>\n",
+       "      <td>snowflake/snowflake-arctic-embed-m</td>\n",
+       "      <td>768</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 512...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.430</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>14</th>\n",
+       "      <td>nomic-ai/nomic-embed-text-v1.5</td>\n",
+       "      <td>768</td>\n",
+       "      <td>Text embeddings, Multimodal (text, image), Eng...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.520</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>15</th>\n",
+       "      <td>jinaai/jina-embeddings-v2-base-en</td>\n",
+       "      <td>768</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 819...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.520</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>16</th>\n",
+       "      <td>nomic-ai/nomic-embed-text-v1</td>\n",
+       "      <td>768</td>\n",
+       "      <td>Text embeddings, Multimodal (text, image), Eng...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.520</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>17</th>\n",
+       "      <td>snowflake/snowflake-arctic-embed-m-long</td>\n",
+       "      <td>768</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 204...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.540</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>18</th>\n",
+       "      <td>mixedbread-ai/mxbai-embed-large-v1</td>\n",
+       "      <td>1024</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 512...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.640</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>19</th>\n",
+       "      <td>jinaai/jina-embeddings-v2-base-code</td>\n",
+       "      <td>768</td>\n",
+       "      <td>Text embeddings, Unimodal (text), Multilingual...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.640</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>20</th>\n",
+       "      <td>sentence-transformers/paraphrase-multilingual-...</td>\n",
+       "      <td>768</td>\n",
+       "      <td>Text embeddings, Unimodal (text), Multilingual...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>1.000</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>21</th>\n",
+       "      <td>snowflake/snowflake-arctic-embed-l</td>\n",
+       "      <td>1024</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 512...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>1.020</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>22</th>\n",
+       "      <td>thenlper/gte-large</td>\n",
+       "      <td>1024</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 512...</td>\n",
+       "      <td>mit</td>\n",
+       "      <td>1.200</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>23</th>\n",
+       "      <td>BAAI/bge-large-en-v1.5</td>\n",
+       "      <td>1024</td>\n",
+       "      <td>Text embeddings, Unimodal (text), English, 512...</td>\n",
+       "      <td>mit</td>\n",
+       "      <td>1.200</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>24</th>\n",
+       "      <td>intfloat/multilingual-e5-large</td>\n",
+       "      <td>1024</td>\n",
+       "      <td>Text embeddings, Unimodal (text), Multilingual...</td>\n",
+       "      <td>mit</td>\n",
+       "      <td>2.240</td>\n",
+       "    </tr>\n",
+       "  </tbody>\n",
+       "</table>\n",
+       "</div>"
+      ]
+     },
+     "execution_count": 12,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "execution_count": 12
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Supported Sparse Text Embedding Models"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-11-13T09:01:07.038954Z",
+     "start_time": "2024-11-13T09:01:07.019656Z"
+    }
+   },
+   "source": [
+    "(\n",
+    "    pd.DataFrame(SparseTextEmbedding.list_supported_models())\n",
+    "    .sort_values(\"size_in_GB\")\n",
+    "    .drop(columns=[\"sources\", \"model_file\", \"additional_files\"])\n",
+    "    .reset_index(drop=True)\n",
+    ")"
+   ],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "                                     model  vocab_size  \\\n",
+       "0                              Qdrant/bm25         NaN   \n",
+       "1  Qdrant/bm42-all-minilm-l6-v2-attentions     30522.0   \n",
+       "2               prithivida/Splade_PP_en_v1     30522.0   \n",
+       "3                prithvida/Splade_PP_en_v1     30522.0   \n",
+       "\n",
+       "                                         description     license  size_in_GB  \\\n",
+       "0  BM25 as sparse embeddings meant to be used wit...  apache-2.0       0.010   \n",
+       "1  Light sparse embedding model, which assigns an...  apache-2.0       0.090   \n",
+       "2  Independent Implementation of SPLADE++ Model f...  apache-2.0       0.532   \n",
+       "3  Independent Implementation of SPLADE++ Model f...  apache-2.0       0.532   \n",
+       "\n",
+       "  requires_idf  \n",
+       "0         True  \n",
+       "1         True  \n",
+       "2          NaN  \n",
+       "3          NaN  "
+      ],
+      "text/html": [
+       "<div>\n",
+       "<style scoped>\n",
+       "    .dataframe tbody tr th:only-of-type {\n",
+       "        vertical-align: middle;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe tbody tr th {\n",
+       "        vertical-align: top;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe thead th {\n",
+       "        text-align: right;\n",
+       "    }\n",
+       "</style>\n",
+       "<table border=\"1\" class=\"dataframe\">\n",
+       "  <thead>\n",
+       "    <tr style=\"text-align: right;\">\n",
+       "      <th></th>\n",
+       "      <th>model</th>\n",
+       "      <th>vocab_size</th>\n",
+       "      <th>description</th>\n",
+       "      <th>license</th>\n",
+       "      <th>size_in_GB</th>\n",
+       "      <th>requires_idf</th>\n",
+       "    </tr>\n",
+       "  </thead>\n",
+       "  <tbody>\n",
+       "    <tr>\n",
+       "      <th>0</th>\n",
+       "      <td>Qdrant/bm25</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>BM25 as sparse embeddings meant to be used wit...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.010</td>\n",
+       "      <td>True</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>1</th>\n",
+       "      <td>Qdrant/bm42-all-minilm-l6-v2-attentions</td>\n",
+       "      <td>30522.0</td>\n",
+       "      <td>Light sparse embedding model, which assigns an...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.090</td>\n",
+       "      <td>True</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>2</th>\n",
+       "      <td>prithivida/Splade_PP_en_v1</td>\n",
+       "      <td>30522.0</td>\n",
+       "      <td>Independent Implementation of SPLADE++ Model f...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.532</td>\n",
+       "      <td>NaN</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>3</th>\n",
+       "      <td>prithvida/Splade_PP_en_v1</td>\n",
+       "      <td>30522.0</td>\n",
+       "      <td>Independent Implementation of SPLADE++ Model f...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.532</td>\n",
+       "      <td>NaN</td>\n",
+       "    </tr>\n",
+       "  </tbody>\n",
+       "</table>\n",
+       "</div>"
+      ]
+     },
+     "execution_count": 13,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "execution_count": 13
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "collapsed": false
+   },
+   "source": [
+    "## Supported Late Interaction Text Embedding Models"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "metadata": {
+    "collapsed": false,
+    "ExecuteTime": {
+     "end_time": "2024-11-13T09:01:08.074442Z",
+     "start_time": "2024-11-13T09:01:08.056138Z"
+    }
+   },
+   "source": [
+    "(\n",
+    "    pd.DataFrame(LateInteractionTextEmbedding.list_supported_models())\n",
+    "    .sort_values(\"size_in_GB\")\n",
+    "    .drop(columns=[\"sources\", \"model_file\"])\n",
+    "    .reset_index(drop=True)\n",
+    ")"
+   ],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "                                   model  dim  \\\n",
+       "0  answerdotai/answerai-colbert-small-v1   96   \n",
+       "1                 colbert-ir/colbertv2.0  128   \n",
+       "2                 jinaai/jina-colbert-v2  128   \n",
+       "\n",
+       "                                         description       license  \\\n",
+       "0  Text embeddings, Unimodal (text), Multilingual...    apache-2.0   \n",
+       "1                             Late interaction model           mit   \n",
+       "2  New model that expands capabilities of colbert...  cc-by-nc-4.0   \n",
+       "\n",
+       "   size_in_GB        additional_files  \n",
+       "0        0.13                     NaN  \n",
+       "1        0.44                     NaN  \n",
+       "2        2.24  [onnx/model.onnx_data]  "
+      ],
+      "text/html": [
+       "<div>\n",
+       "<style scoped>\n",
+       "    .dataframe tbody tr th:only-of-type {\n",
+       "        vertical-align: middle;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe tbody tr th {\n",
+       "        vertical-align: top;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe thead th {\n",
+       "        text-align: right;\n",
+       "    }\n",
+       "</style>\n",
+       "<table border=\"1\" class=\"dataframe\">\n",
+       "  <thead>\n",
+       "    <tr style=\"text-align: right;\">\n",
+       "      <th></th>\n",
+       "      <th>model</th>\n",
+       "      <th>dim</th>\n",
+       "      <th>description</th>\n",
+       "      <th>license</th>\n",
+       "      <th>size_in_GB</th>\n",
+       "      <th>additional_files</th>\n",
+       "    </tr>\n",
+       "  </thead>\n",
+       "  <tbody>\n",
+       "    <tr>\n",
+       "      <th>0</th>\n",
+       "      <td>answerdotai/answerai-colbert-small-v1</td>\n",
+       "      <td>96</td>\n",
+       "      <td>Text embeddings, Unimodal (text), Multilingual...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.13</td>\n",
+       "      <td>NaN</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>1</th>\n",
+       "      <td>colbert-ir/colbertv2.0</td>\n",
+       "      <td>128</td>\n",
+       "      <td>Late interaction model</td>\n",
+       "      <td>mit</td>\n",
+       "      <td>0.44</td>\n",
+       "      <td>NaN</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>2</th>\n",
+       "      <td>jinaai/jina-colbert-v2</td>\n",
+       "      <td>128</td>\n",
+       "      <td>New model that expands capabilities of colbert...</td>\n",
+       "      <td>cc-by-nc-4.0</td>\n",
+       "      <td>2.24</td>\n",
+       "      <td>[onnx/model.onnx_data]</td>\n",
+       "    </tr>\n",
+       "  </tbody>\n",
+       "</table>\n",
+       "</div>"
+      ]
+     },
+     "execution_count": 14,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "execution_count": 14
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "collapsed": false
+   },
+   "source": [
+    "## Supported Image Embedding Models"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "metadata": {
+    "collapsed": false,
+    "ExecuteTime": {
+     "end_time": "2024-11-13T09:01:09.171647Z",
+     "start_time": "2024-11-13T09:01:09.150940Z"
+    }
+   },
+   "source": [
+    "(\n",
+    "    pd.DataFrame(ImageEmbedding.list_supported_models())\n",
+    "    .sort_values(\"size_in_GB\")\n",
+    "    .drop(columns=[\"sources\", \"model_file\"])\n",
+    "    .reset_index(drop=True)\n",
+    ")"
+   ],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "                         model   dim  \\\n",
+       "0         Qdrant/resnet50-onnx  2048   \n",
+       "1  Qdrant/clip-ViT-B-32-vision   512   \n",
+       "2       Qdrant/Unicom-ViT-B-32   512   \n",
+       "3       Qdrant/Unicom-ViT-B-16   768   \n",
+       "\n",
+       "                                         description     license  size_in_GB  \n",
+       "0      Image embeddings, Unimodal (image), 2016 year  apache-2.0        0.10  \n",
+       "1  Image embeddings, Multimodal (text&image), 202...         mit        0.34  \n",
+       "2  Image embeddings, Multimodal (text&image), 202...  apache-2.0        0.48  \n",
+       "3  Image embeddings (more detailed than Unicom-Vi...  apache-2.0        0.82  "
+      ],
+      "text/html": [
+       "<div>\n",
+       "<style scoped>\n",
+       "    .dataframe tbody tr th:only-of-type {\n",
+       "        vertical-align: middle;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe tbody tr th {\n",
+       "        vertical-align: top;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe thead th {\n",
+       "        text-align: right;\n",
+       "    }\n",
+       "</style>\n",
+       "<table border=\"1\" class=\"dataframe\">\n",
+       "  <thead>\n",
+       "    <tr style=\"text-align: right;\">\n",
+       "      <th></th>\n",
+       "      <th>model</th>\n",
+       "      <th>dim</th>\n",
+       "      <th>description</th>\n",
+       "      <th>license</th>\n",
+       "      <th>size_in_GB</th>\n",
+       "    </tr>\n",
+       "  </thead>\n",
+       "  <tbody>\n",
+       "    <tr>\n",
+       "      <th>0</th>\n",
+       "      <td>Qdrant/resnet50-onnx</td>\n",
+       "      <td>2048</td>\n",
+       "      <td>Image embeddings, Unimodal (image), 2016 year</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.10</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>1</th>\n",
+       "      <td>Qdrant/clip-ViT-B-32-vision</td>\n",
+       "      <td>512</td>\n",
+       "      <td>Image embeddings, Multimodal (text&amp;image), 202...</td>\n",
+       "      <td>mit</td>\n",
+       "      <td>0.34</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>2</th>\n",
+       "      <td>Qdrant/Unicom-ViT-B-32</td>\n",
+       "      <td>512</td>\n",
+       "      <td>Image embeddings, Multimodal (text&amp;image), 202...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.48</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>3</th>\n",
+       "      <td>Qdrant/Unicom-ViT-B-16</td>\n",
+       "      <td>768</td>\n",
+       "      <td>Image embeddings (more detailed than Unicom-Vi...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "      <td>0.82</td>\n",
+       "    </tr>\n",
+       "  </tbody>\n",
+       "</table>\n",
+       "</div>"
+      ]
+     },
+     "execution_count": 15,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "execution_count": 15
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Supported Rerank Cross Encoder Models"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-11-13T09:01:10.313943Z",
+     "start_time": "2024-11-13T09:01:10.298428Z"
+    }
+   },
+   "source": [
+    "(\n",
+    "    pd.DataFrame(TextCrossEncoder.list_supported_models())\n",
+    "    .sort_values(\"size_in_GB\")\n",
+    "    .drop(columns=[\"sources\", \"model_file\"])\n",
+    "    .reset_index(drop=True)\n",
+    ")"
+   ],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "                                       model  size_in_GB  \\\n",
+       "0              Xenova/ms-marco-MiniLM-L-6-v2        0.08   \n",
+       "1             Xenova/ms-marco-MiniLM-L-12-v2        0.12   \n",
+       "2            jinaai/jina-reranker-v1-tiny-en        0.13   \n",
+       "3           jinaai/jina-reranker-v1-turbo-en        0.15   \n",
+       "4                     BAAI/bge-reranker-base        1.04   \n",
+       "5  jinaai/jina-reranker-v2-base-multilingual        1.11   \n",
+       "\n",
+       "                                         description       license  \n",
+       "0  MiniLM-L-6-v2 model optimized for re-ranking t...    apache-2.0  \n",
+       "1  MiniLM-L-12-v2 model optimized for re-ranking ...    apache-2.0  \n",
+       "2  Designed for blazing-fast re-ranking with 8K c...    apache-2.0  \n",
+       "3  Designed for blazing-fast re-ranking with 8K c...    apache-2.0  \n",
+       "4  BGE reranker base model for cross-encoder re-r...           mit  \n",
+       "5  A multi-lingual reranker model for cross-encod...  cc-by-nc-4.0  "
+      ],
+      "text/html": [
+       "<div>\n",
+       "<style scoped>\n",
+       "    .dataframe tbody tr th:only-of-type {\n",
+       "        vertical-align: middle;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe tbody tr th {\n",
+       "        vertical-align: top;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe thead th {\n",
+       "        text-align: right;\n",
+       "    }\n",
+       "</style>\n",
+       "<table border=\"1\" class=\"dataframe\">\n",
+       "  <thead>\n",
+       "    <tr style=\"text-align: right;\">\n",
+       "      <th></th>\n",
+       "      <th>model</th>\n",
+       "      <th>size_in_GB</th>\n",
+       "      <th>description</th>\n",
+       "      <th>license</th>\n",
+       "    </tr>\n",
+       "  </thead>\n",
+       "  <tbody>\n",
+       "    <tr>\n",
+       "      <th>0</th>\n",
+       "      <td>Xenova/ms-marco-MiniLM-L-6-v2</td>\n",
+       "      <td>0.08</td>\n",
+       "      <td>MiniLM-L-6-v2 model optimized for re-ranking t...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>1</th>\n",
+       "      <td>Xenova/ms-marco-MiniLM-L-12-v2</td>\n",
+       "      <td>0.12</td>\n",
+       "      <td>MiniLM-L-12-v2 model optimized for re-ranking ...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>2</th>\n",
+       "      <td>jinaai/jina-reranker-v1-tiny-en</td>\n",
+       "      <td>0.13</td>\n",
+       "      <td>Designed for blazing-fast re-ranking with 8K c...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>3</th>\n",
+       "      <td>jinaai/jina-reranker-v1-turbo-en</td>\n",
+       "      <td>0.15</td>\n",
+       "      <td>Designed for blazing-fast re-ranking with 8K c...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>4</th>\n",
+       "      <td>BAAI/bge-reranker-base</td>\n",
+       "      <td>1.04</td>\n",
+       "      <td>BGE reranker base model for cross-encoder re-r...</td>\n",
+       "      <td>mit</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>5</th>\n",
+       "      <td>jinaai/jina-reranker-v2-base-multilingual</td>\n",
+       "      <td>1.11</td>\n",
+       "      <td>A multi-lingual reranker model for cross-encod...</td>\n",
+       "      <td>cc-by-nc-4.0</td>\n",
+       "    </tr>\n",
+       "  </tbody>\n",
+       "</table>\n",
+       "</div>"
+      ]
+     },
+     "execution_count": 16,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "execution_count": 16
+  },
+  {
+   "metadata": {},
+   "cell_type": "code",
+   "outputs": [],
+   "execution_count": null,
+   "source": ""
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3.8.18 ('base')",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.11.8"
+  },
+  "orig_nbformat": 4,
+  "vscode": {
+   "interpreter": {
+    "hash": "c4a27af61e455bc18dcf16f5867a2ff0402fa12b01dd0f6ce3a79ae73ad15e91"
+   }
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}
@@ -3,22 +3,7 @@
  {
   "cell_type": "markdown",
   "metadata": {},
-   "source": [
-    "# Binary Quantization of OpenAI Embedding\n",
-    "---\n",
-    "\n",
-    "In the world of large-scale data retrieval and processing, efficiency is crucial. With the exponential growth of data, the ability to retrieve information quickly and accurately can significantly affect system performance. This blog post explores a technique known as binary quantization applied to OpenAI embeddings, demonstrating how it can enhance **retrieval latency by 20x** or more.\n",
-    "\n",
-    "## What Are OpenAI Embeddings?\n",
-    "OpenAI embeddings are numerical representations of textual information. They transform text into a vector space where semantically similar texts are mapped close together. This mathematical representation enables computers to understand and process human language more effectively.\n",
-    "\n",
-    "## Binary Quantization\n",
-    "Binary quantization is a method which converts continuous numerical values into binary values (0 or 1). It simplifies the data structure, allowing faster computations. Here's a brief overview of the binary quantization process applied to OpenAI embeddings:\n",
-    "\n",
-    "1. **Load Embeddings**: OpenAI embeddings are loaded from parquet files.\n",
-    "2. **Binary Transformation**: The continuous valued vectors are converted into binary form. Here, values greater than 0 are set to 1, and others remain 0.\n",
-    "3. **Comparison & Retrieval**: Binary vectors are used for comparison using logical XOR operations and other efficient algorithms."
-   ]
+   "source": []
  },
  {
   "cell_type": "markdown",
@@ -29,24 +14,33 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 11,
-   "metadata": {},
+   "execution_count": 1,
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-06T17:00:06.460001Z",
+     "start_time": "2024-06-06T17:00:04.214098Z"
+    }
+   },
   "outputs": [],
   "source": [
-    "!pip install matplotlib tqdm pandas numpy --quiet"
+    "!pip install matplotlib tqdm pandas numpy datasets --quiet --upgrade"
   ]
  },
  {
   "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": 2,
   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-06T17:00:07.041784Z",
+     "start_time": "2024-06-06T17:00:06.461658Z"
+    },
    "id": "WBVTItUX4yyr"
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
-    "import matplotlib.pyplot as plt\n",
+    "from datasets import load_dataset\n",
    "from tqdm import tqdm"
   ]
  },
@@ -68,8 +62,12 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 13,
+   "execution_count": 3,
   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-06T17:01:09.343230Z",
+     "start_time": "2024-06-06T17:00:07.042526Z"
+    },
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 250
@@ -77,58 +75,24 @@
    "id": "REJpFqkG7EG2",
    "outputId": "7a43c0ae-fbcc-45fe-fd58-bfe691297b22"
   },
-   "outputs": [
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "100%|██████████| 26/26 [00:10<00:00,  2.45it/s]\n"
-     ]
-    },
-    {
-     "data": {
-      "text/plain": [
-       "(1000000, 1536)"
-      ]
-     },
-     "execution_count": 13,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
+   "outputs": [],
   "source": [
-    "def get_openai_vectors(force_download: bool = False):\n",
-    "    res = []\n",
-    "    for i in tqdm(range(26)):\n",
-    "        if force_download:\n",
-    "            !wget https://huggingface.co/api/datasets/KShivendu/dbpedia-entities-openai-1M/parquet/KShivendu--dbpedia-entities-openai-1M/train/{i}.parquet\n",
-    "        df = pd.read_parquet(f\"{i}.parquet\", engine=\"pyarrow\")\n",
-    "        res.append(np.stack(df.openai))\n",
-    "        del df\n",
-    "\n",
-    "    openai_vectors = np.concatenate(res)\n",
-    "    del res\n",
-    "    return openai_vectors\n",
-    "\n",
-    "\n",
-    "openai_vectors = get_openai_vectors(force_download=False)\n",
-    "openai_vectors.shape"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## ㆓ Binary Conversion\n",
-    "\n",
-    "Here, we will use 0 as the threshold for the binary conversion. All values greater than 0 will be set to 1, and others will remain 0. This is a simple and effective way to convert continuous values into binary values for OpenAI embeddings."
+    "# Download from Huggingface Hub\n",
+    "ds = load_dataset(\n",
+    "    \"Qdrant/dbpedia-entities-openai3-text-embedding-3-large-3072-100K\", split=\"train\"\n",
+    ")\n",
+    "openai_vectors = np.array(ds[\"text-embedding-3-large-3072-embedding\"])\n",
+    "del ds"
   ]
  },
  {
   "cell_type": "code",
-   "execution_count": 14,
+   "execution_count": 4,
   "metadata": {
-    "id": "0JM2-Bj2Jkab"
+    "ExecuteTime": {
+     "end_time": "2024-06-06T17:01:10.900963Z",
+     "start_time": "2024-06-06T17:01:09.344842Z"
+    }
   },
   "outputs": [],
   "source": [
@@ -136,6 +100,30 @@
    "openai_bin[openai_vectors > 0] = 1"
   ]
  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-06T17:01:10.906827Z",
+     "start_time": "2024-06-06T17:01:10.901820Z"
+    }
+   },
+   "outputs": [
+    {
+     "data": {
+      "text/plain": "3072"
+     },
+     "execution_count": 5,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "n_dim = openai_vectors.shape[1]\n",
+    "n_dim"
+   ]
+  },
  {
   "cell_type": "markdown",
   "metadata": {},
@@ -147,8 +135,12 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 15,
+   "execution_count": 6,
   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-06T17:01:10.909730Z",
+     "start_time": "2024-06-06T17:01:10.908166Z"
+    },
    "id": "FqshI-GlIERd"
   },
   "outputs": [],
@@ -157,7 +149,7 @@
    "    scores = np.dot(openai_vectors, openai_vectors[idx])\n",
    "    dot_results = np.argsort(scores)[-limit:][::-1]\n",
    "\n",
-    "    bin_scores = 1536 - np.logical_xor(openai_bin, openai_bin[idx]).sum(axis=1)\n",
+    "    bin_scores = n_dim - np.logical_xor(openai_bin, openai_bin[idx]).sum(axis=1)\n",
    "    bin_results = np.argsort(bin_scores)[-(limit * oversampling) :][::-1]\n",
    "\n",
    "    return len(set(dot_results).intersection(set(bin_results))) / limit"
@@ -172,8 +164,12 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 18,
+   "execution_count": 7,
   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-06T17:01:25.206592Z",
+     "start_time": "2024-06-06T17:01:10.911971Z"
+    },
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
@@ -185,110 +181,128 @@
     "name": "stderr",
     "output_type": "stream",
     "text": [
-      "  0%|          | 0/4 [00:00<?, ?it/s]"
+      "  0%|          | 0/4 [00:00<?, ?it/s]\n",
+      "  0%|          | 0/2 [00:00<?, ?it/s]\u001b[A\n",
+      " 50%|█████     | 1/2 [00:02<00:02,  2.05s/it]\u001b[A"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
-      "{'sampling_rate': 1, 'limit': 10, 'recall': 0.8}\n"
+      "{'sampling_rate': 1, 'limit': 3, 'mean_acc': 0.9}\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
-      "100%|██████████| 2/2 [00:33<00:00, 16.98s/it]\n",
-      " 25%|██▌       | 1/4 [00:33<01:41, 33.96s/it]"
+      "\n",
+      "100%|██████████| 2/2 [00:04<00:00,  2.02s/it]\u001b[A\n",
+      " 25%|██▌       | 1/4 [00:04<00:12,  4.05s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
-      "{'sampling_rate': 1, 'limit': 100, 'recall': 0.708}\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": []
-    },
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "{'sampling_rate': 2, 'limit': 10, 'recall': 0.95}\n"
+      "{'sampling_rate': 1, 'limit': 10, 'mean_acc': 0.8300000000000001}\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
-      "100%|██████████| 2/2 [00:32<00:00, 16.38s/it]\n",
-      " 50%|█████     | 2/4 [01:06<01:06, 33.26s/it]"
+      "\n",
+      "  0%|          | 0/2 [00:00<?, ?it/s]\u001b[A\n",
+      " 50%|█████     | 1/2 [00:01<00:01,  1.72s/it]\u001b[A"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
-      "{'sampling_rate': 2, 'limit': 100, 'recall': 0.877}\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": []
-    },
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "{'sampling_rate': 3, 'limit': 10, 'recall': 0.96}\n"
+      "{'sampling_rate': 2, 'limit': 3, 'mean_acc': 1.0}\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
-      "100%|██████████| 2/2 [00:32<00:00, 16.49s/it]\n",
-      " 75%|███████▌  | 3/4 [01:39<00:33, 33.13s/it]"
+      "\n",
+      "100%|██████████| 2/2 [00:03<00:00,  1.76s/it]\u001b[A\n",
+      " 50%|█████     | 2/4 [00:07<00:07,  3.75s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
-      "{'sampling_rate': 3, 'limit': 100, 'recall': 0.937}\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": []
-    },
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "{'sampling_rate': 5, 'limit': 10, 'recall': 0.9800000000000001}\n"
+      "{'sampling_rate': 2, 'limit': 10, 'mean_acc': 0.9700000000000001}\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
-      "100%|██████████| 2/2 [00:32<00:00, 16.47s/it]\n",
-      "100%|██████████| 4/4 [02:12<00:00, 33.17s/it]"
+      "\n",
+      "  0%|          | 0/2 [00:00<?, ?it/s]\u001b[A\n",
+      " 50%|█████     | 1/2 [00:01<00:01,  1.72s/it]\u001b[A"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
-      "{'sampling_rate': 5, 'limit': 100, 'recall': 0.977}\n"
+      "{'sampling_rate': 3, 'limit': 3, 'mean_acc': 1.0}\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "100%|██████████| 2/2 [00:03<00:00,  1.69s/it]\u001b[A\n",
+      " 75%|███████▌  | 3/4 [00:10<00:03,  3.58s/it]"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "{'sampling_rate': 3, 'limit': 10, 'mean_acc': 0.9800000000000001}\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "  0%|          | 0/2 [00:00<?, ?it/s]\u001b[A\n",
+      " 50%|█████     | 1/2 [00:01<00:01,  1.68s/it]\u001b[A"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "{'sampling_rate': 5, 'limit': 3, 'mean_acc': 1.0}\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "100%|██████████| 2/2 [00:03<00:00,  1.65s/it]\u001b[A\n",
+      "100%|██████████| 4/4 [00:14<00:00,  3.57s/it]"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "{'sampling_rate': 5, 'limit': 10, 'mean_acc': 0.99}\n"
     ]
    },
    {
@@ -301,117 +315,53 @@
   ],
   "source": [
    "number_of_samples = 10\n",
-    "limits = [10, 100]\n",
+    "limits = [3, 10]\n",
    "sampling_rate = [1, 2, 3, 5]\n",
    "results = []\n",
    "\n",
+    "\n",
    "def mean_accuracy(number_of_samples, limit, sampling_rate):\n",
-    "    return np.mean([accuracy(i, limit=limit, oversampling=sampling_rate) for i in range(number_of_samples)])\n",
+    "    return np.mean(\n",
+    "        [accuracy(i, limit=limit, oversampling=sampling_rate) for i in range(number_of_samples)]\n",
+    "    )\n",
+    "\n",
    "\n",
    "for i in tqdm(sampling_rate):\n",
    "    for j in tqdm(limits):\n",
-    "        result = {\"sampling_rate\": i, \"limit\": j, \"recall\": mean_accuracy(number_of_samples, j, i)}\n",
+    "        result = {\n",
+    "            \"sampling_rate\": i,\n",
+    "            \"limit\": j,\n",
+    "            \"mean_acc\": mean_accuracy(number_of_samples, j, i),\n",
+    "        }\n",
    "        print(result)\n",
    "        results.append(result)"
   ]
  },
  {
-   "cell_type": "code",
-   "execution_count": 19,
+   "cell_type": "markdown",
   "metadata": {},
+   "source": [
+    "## ㆓ Binary Conversion\n",
+    "\n",
+    "Here, we will use 0 as the threshold for the binary conversion. All values greater than 0 will be set to 1, and others will remain 0. This is a simple and effective way to convert continuous values into binary values for OpenAI embeddings."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-06-06T17:01:25.247495Z",
+     "start_time": "2024-06-06T17:01:25.213508Z"
+    }
+   },
   "outputs": [
    {
     "data": {
-      "text/html": [
-       "<div>\n",
-       "<style scoped>\n",
-       "    .dataframe tbody tr th:only-of-type {\n",
-       "        vertical-align: middle;\n",
-       "    }\n",
-       "\n",
-       "    .dataframe tbody tr th {\n",
-       "        vertical-align: top;\n",
-       "    }\n",
-       "\n",
-       "    .dataframe thead th {\n",
-       "        text-align: right;\n",
-       "    }\n",
-       "</style>\n",
-       "<table border=\"1\" class=\"dataframe\">\n",
-       "  <thead>\n",
-       "    <tr style=\"text-align: right;\">\n",
-       "      <th></th>\n",
-       "      <th>sampling_rate</th>\n",
-       "      <th>limit</th>\n",
-       "      <th>recall</th>\n",
-       "    </tr>\n",
-       "  </thead>\n",
-       "  <tbody>\n",
-       "    <tr>\n",
-       "      <th>0</th>\n",
-       "      <td>1</td>\n",
-       "      <td>10</td>\n",
-       "      <td>0.800</td>\n",
-       "    </tr>\n",
-       "    <tr>\n",
-       "      <th>1</th>\n",
-       "      <td>1</td>\n",
-       "      <td>100</td>\n",
-       "      <td>0.708</td>\n",
-       "    </tr>\n",
-       "    <tr>\n",
-       "      <th>2</th>\n",
-       "      <td>2</td>\n",
-       "      <td>10</td>\n",
-       "      <td>0.950</td>\n",
-       "    </tr>\n",
-       "    <tr>\n",
-       "      <th>3</th>\n",
-       "      <td>2</td>\n",
-       "      <td>100</td>\n",
-       "      <td>0.877</td>\n",
-       "    </tr>\n",
-       "    <tr>\n",
-       "      <th>4</th>\n",
-       "      <td>3</td>\n",
-       "      <td>10</td>\n",
-       "      <td>0.960</td>\n",
-       "    </tr>\n",
-       "    <tr>\n",
-       "      <th>5</th>\n",
-       "      <td>3</td>\n",
-       "      <td>100</td>\n",
-       "      <td>0.937</td>\n",
-       "    </tr>\n",
-       "    <tr>\n",
-       "      <th>6</th>\n",
-       "      <td>5</td>\n",
-       "      <td>10</td>\n",
-       "      <td>0.980</td>\n",
-       "    </tr>\n",
-       "    <tr>\n",
-       "      <th>7</th>\n",
-       "      <td>5</td>\n",
-       "      <td>100</td>\n",
-       "      <td>0.977</td>\n",
-       "    </tr>\n",
-       "  </tbody>\n",
-       "</table>\n",
-       "</div>"
-      ],
-      "text/plain": [
-       "   sampling_rate  limit  recall\n",
-       "0              1     10   0.800\n",
-       "1              1    100   0.708\n",
-       "2              2     10   0.950\n",
-       "3              2    100   0.877\n",
-       "4              3     10   0.960\n",
-       "5              3    100   0.937\n",
-       "6              5     10   0.980\n",
-       "7              5    100   0.977"
-      ]
+      "text/html": "<div>\n<style scoped>\n    .dataframe tbody tr th:only-of-type {\n        vertical-align: middle;\n    }\n\n    .dataframe tbody tr th {\n        vertical-align: top;\n    }\n\n    .dataframe thead th {\n        text-align: right;\n    }\n</style>\n<table border=\"1\" class=\"dataframe\">\n  <thead>\n    <tr style=\"text-align: right;\">\n      <th></th>\n      <th>sampling_rate</th>\n      <th>limit</th>\n      <th>mean_acc</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>0</th>\n      <td>1</td>\n      <td>3</td>\n      <td>0.90</td>\n    </tr>\n    <tr>\n      <th>1</th>\n      <td>1</td>\n      <td>10</td>\n      <td>0.83</td>\n    </tr>\n    <tr>\n      <th>2</th>\n      <td>2</td>\n      <td>3</td>\n      <td>1.00</td>\n    </tr>\n    <tr>\n      <th>3</th>\n      <td>2</td>\n      <td>10</td>\n      <td>0.97</td>\n    </tr>\n    <tr>\n      <th>4</th>\n      <td>3</td>\n      <td>3</td>\n      <td>1.00</td>\n    </tr>\n    <tr>\n      <th>5</th>\n      <td>3</td>\n      <td>10</td>\n      <td>0.98</td>\n    </tr>\n    <tr>\n      <th>6</th>\n      <td>5</td>\n      <td>3</td>\n      <td>1.00</td>\n    </tr>\n    <tr>\n      <th>7</th>\n      <td>5</td>\n      <td>10</td>\n      <td>0.99</td>\n    </tr>\n  </tbody>\n</table>\n</div>",
+      "text/plain": "   sampling_rate  limit  mean_acc\n0              1      3      0.90\n1              1     10      0.83\n2              2      3      1.00\n3              2     10      0.97\n4              3      3      1.00\n5              3     10      0.98\n6              5      3      1.00\n7              5     10      0.99"
     },
-     "execution_count": 19,
+     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
@@ -422,22 +372,13 @@
   ]
  },
  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "| sampling_rate | limit | accuracy |\n",
-    "|---------------|-------|----------|\n",
-    "| 1             | 10    | 0.800    |\n",
-    "| 1             | 100   | 0.708    |\n",
-    "| 2             | 10    | 0.950    |\n",
-    "| 2             | 100   | 0.877    |\n",
-    "| 4             | 10    | 0.970    |\n",
-    "| 4             | 100   | 0.956    |\n",
-    "| 8             | 10    | 0.990    |\n",
-    "| 8             | 100   | 0.990    |\n",
-    "| 16            | 10    | 1.000    |\n",
-    "| 16            | 100   | 0.998    |"
-   ]
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [],
+   "source": []
  }
 ],
 "metadata": {
@@ -446,7 +387,8 @@
   "provenance": []
  },
  "kernelspec": {
-   "display_name": "Python 3",
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
   "name": "python3"
  },
  "language_info": {
@@ -459,7 +401,7 @@
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
-   "version": "3.9.17"
+   "version": "3.10.13"
  }
 },
 "nbformat": 4,
@@ -1,12 +1,21 @@
 # ⚡️ What is FastEmbed?

-FastEmbed is an easy to use -- lightweight, fast, Python library built for retrieval embedding generation. 
+FastEmbed is a lightweight, fast, Python library built for embedding generation. We [support popular text models](https://qdrant.github.io/fastembed/examples/Supported_Models/). Please [open a Github issue](https://github.com/qdrant/fastembed/issues/new) if you want us to add a new model.

-The default embedding supports "query" and "passage" prefixes for the input text. The default model is [Flag Embedding](https://github.com/FlagOpen/FlagEmbedding), which is top of the [MTEB](https://huggingface.co/spaces/mteb/leaderboard) leaderboard. 
+1. Light & Fast
+    - Quantized model weights
+    - ONNX Runtime for inference

-Advanced user? Skip ahead to [Retrieval with FastEmbed](https://qdrant.github.io/fastembed/examples/Retrieval_with_FastEmbed/)
+2. Accuracy/Recall
+    - Better than OpenAI Ada-002
+    - Default is Flag Embedding, which has shown good results on the [MTEB](https://huggingface.co/spaces/mteb/leaderboard) leaderboard
+    - List of [supported models](https://qdrant.github.io/fastembed/examples/Supported_Models/) - including multilingual models

-To install the FastEmbed library, pip works: 
+Here is an example for [Retrieval Embedding Generation](https://qdrant.github.io/fastembed/examples/Retrieval%20with%20FastEmbed/) and how to use [FastEmbed with Qdrant](https://qdrant.github.io/fastembed/examples/Usage_With_Qdrant/).
+
+## 🚀 Installation
+
+To install the FastEmbed library, pip works:

 ```bash
 pip install fastembed
@@ -15,31 +24,52 @@ pip install fastembed
 ## 📖 Usage

 ```python
-from fastembed.embedding import DefaultEmbedding
+from fastembed import TextEmbedding

-documents: List[str] = [
+documents: list[str] = [
    "passage: Hello, World!",
-    "query: Hello, World!", # these are two different embedding
+    "query: Hello, World!",
    "passage: This is an example passage.",
-    # You can leave out the prefix but it's recommended
-    "fastembed is supported by and maintained by Qdrant." 
+    "fastembed is supported by and maintained by Qdrant."
 ]
-embedding_model = DefaultEmbedding() 
-embeddings: List[np.ndarray] = list(embedding_model.embed(documents))
+embedding_model = TextEmbedding()
+embeddings: list[np.ndarray] = embedding_model.embed(documents)
 ```

-## 🚒 Under the hood
+## Usage with Qdrant

-### Why fast?
+Installation with Qdrant Client in Python:

-It's important we justify the "fast" in FastEmbed. FastEmbed is fast because:
+```bash
+pip install qdrant-client[fastembed]
+```

-1. Quantized model weights
-2. ONNX Runtime which allows for inference on CPU, GPU, and other dedicated runtimes
+Might have to use ```pip install 'qdrant-client[fastembed]'``` on zsh.

-### Why light?
-1. No hidden dependencies on PyTorch or TensorFlow via Huggingface Transformers
+```python
+from qdrant_client import QdrantClient

-### Why accurate?
-1. Better than OpenAI Ada-002
-2. Top of the Embedding leaderboards e.g. [MTEB](https://huggingface.co/spaces/mteb/leaderboard)
+# Initialize the client
+client = QdrantClient(":memory:")  # Using an in-process Qdrant
+
+# Prepare your documents, metadata, and IDs
+docs = ["Qdrant has Langchain integrations", "Qdrant also has Llama Index integrations"]
+metadata = [
+    {"source": "Langchain-docs"},
+    {"source": "Llama-index-docs"},
+]
+ids = [42, 2]
+
+client.add(
+    collection_name="demo_collection",
+    documents=docs,
+    metadata=metadata,
+    ids=ids
+)
+
+search_result = client.query(
+    collection_name="demo_collection",
+    query_text="This is a query document"
+)
+print(search_result)
+```
@@ -5,7 +5,7 @@
    <a href="{{ page.nb_url }}" title="Download Notebook" class="md-content__button md-icon jp-DownloadNB">
        {% include ".icons/material/download.svg" %}
    </a>
-{% endif %} 
+{% endif %}

 {{ super() }}

@@ -20,7 +20,8 @@
        </span>
        <strong>Qdrant Discord server</strong>
    </a>
-    to get help and share your work! Or check out <a rel="me" href="https://login.cloud.qdrant.io/">Qdrant Cloud</a> to
+    to get help and share your work! Or check out <a rel="me"
+        href="https://cloud.qdrant.io?utm_source=twitter&utm_medium=website&utm_campaign=fastembed">Qdrant Cloud</a> to
    get started with vector search!
 </div>
-{% endblock %}
+{% endblock %}
@@ -21,7 +21,7 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 8,
+   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
@@ -37,13 +37,12 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 9,
+   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
-    "from typing import List\n",
    "import numpy as np\n",
-    "from fastembed.embedding import FlagEmbedding as Embedding"
+    "from fastembed import TextEmbedding"
   ]
  },
  {
@@ -58,7 +57,7 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 10,
+   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
@@ -71,7 +70,7 @@
   ],
   "source": [
    "# Example list of documents\n",
-    "documents: List[str] = [\n",
+    "documents: list[str] = [\n",
    "    \"Maharana Pratap was a Rajput warrior king from Mewar\",\n",
    "    \"He fought against the Mughal Empire led by Akbar\",\n",
    "    \"The Battle of Haldighati in 1576 was his most famous battle\",\n",
@@ -84,10 +83,10 @@
    "    \"His life has been depicted in various films, TV shows, and books\",\n",
    "]\n",
    "# Initialize the DefaultEmbedding class with the desired parameters\n",
-    "embedding_model = Embedding(model_name=\"BAAI/bge-small-en\", max_length=512)\n",
+    "embedding_model = TextEmbedding(model_name=\"BAAI/bge-small-en\")\n",
    "\n",
    "# We'll use the passage_embed method to get the embeddings for the documents\n",
-    "embeddings: List[np.ndarray] = list(\n",
+    "embeddings: list[np.ndarray] = list(\n",
    "    embedding_model.passage_embed(documents)\n",
    ")  # notice that we are casting the generator to a list\n",
    "\n",
@@ -105,7 +104,7 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 11,
+   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
@@ -124,65 +123,27 @@
    "        print(f\"Rank {i+1}: {documents[sorted_scores[i]]}\")"
   ]
  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Running and Comparing Queries\n",
-    "Finally, we run our sample query using the `print_top_k` function.\n",
-    "\n",
-    "The differences between using query embeddings and plain embeddings can be observed in the retrieved ranks:\n",
-    "\n",
-    "Using query embeddings (from `query_embed` method):"
-   ]
-  },
  {
   "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Rank 1: Maharana Pratap was a Rajput warrior king from Mewar\n",
-      "Rank 2: Maharana Pratap is considered a symbol of Rajput resistance against foreign rule\n",
-      "Rank 3: His legacy is celebrated in Rajasthan through festivals and monuments\n",
-      "Rank 4: His capital was Chittorgarh, which he lost to the Mughals\n",
-      "Rank 5: He fought against the Mughal Empire led by Akbar\n"
-     ]
+     "data": {
+      "text/plain": [
+       "(array([-0.06002192,  0.04322132, -0.00545516, -0.04419701, -0.00542277],\n",
+       "       dtype=float32),\n",
+       " array([-0.06002192,  0.04322132, -0.00545516, -0.04419701, -0.00542277],\n",
+       "       dtype=float32))"
+      ]
+     },
+     "execution_count": 5,
+     "metadata": {},
+     "output_type": "execute_result"
    }
   ],
   "source": [
-    "print_top_k(query_embedding, embeddings, documents)"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "Using plain embeddings (from `embed` method):"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 13,
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Rank 1: He died in 1597 at the age of 57\n",
-      "Rank 2: His life has been depicted in various films, TV shows, and books\n",
-      "Rank 3: Maharana Pratap was a Rajput warrior king from Mewar\n",
-      "Rank 4: He had 11 wives and 17 sons, including Amar Singh I who succeeded him as ruler of Mewar\n",
-      "Rank 5: He fought against the Mughal Empire led by Akbar\n"
-     ]
-    }
-   ],
-   "source": [
-    "print_top_k(plain_query_embedding, embeddings, documents)"
+    "query_embedding[:5], plain_query_embedding[:5]"
   ]
  },
  {
@@ -213,7 +174,7 @@
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
-   "version": "3.9.17"
+   "version": "3.10.13"
  },
  "orig_nbformat": 4
 },
@@ -5,9 +5,7 @@
   "cell_type": "markdown",
   "metadata": {},
   "source": [
-    "# [Experimental] Usage With Qdrant\n",
-    "\n",
-    "> **Note:** This notebook is experimental and is subject to change. For working with this, use the dev branch of QdrantClient.\n",
+    "# Usage With Qdrant\n",
    "\n",
    "This notebook demonstrates how to use FastEmbed and Qdrant to perform vector search and retrieval. Qdrant is an open-source vector similarity search engine that is used to store, organize, and query collections of high-dimensional vectors. \n",
    "\n",
@@ -28,13 +26,11 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 14,
+   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
-    "# !pip install fastembed --quiet --upgrade\n",
-    "\n",
-    "# !pip install git+https://github.com/qdrant/qdrant_client.git@dev"
+    "!pip install 'qdrant-client[fastembed]' --quiet --upgrade"
   ]
  },
  {
@@ -46,13 +42,10 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 15,
+   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
-    "from typing import List\n",
-    "import numpy as np\n",
-    "from fastembed.embedding import FlagEmbedding as Embedding\n",
    "from qdrant_client import QdrantClient"
   ]
  },
@@ -68,12 +61,12 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 16,
+   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Example list of documents\n",
-    "documents: List[str] = [\n",
+    "documents: list[str] = [\n",
    "    \"Maharana Pratap was a Rajput warrior king from Mewar\",\n",
    "    \"He fought against the Mughal Empire led by Akbar\",\n",
    "    \"The Battle of Haldighati in 1576 was his most famous battle\",\n",
@@ -105,25 +98,32 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 17,
+   "execution_count": 4,
   "metadata": {},
   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "100%|██████████| 77.7M/77.7M [00:05<00:00, 14.6MiB/s]\n"
+     ]
+    },
    {
     "data": {
      "text/plain": [
-       "['901ff9d7a90a4e56afe655d1de8e7c06',\n",
-       " '7bae5aa894164398a9be68d47d72ff7a',\n",
-       " 'd940bba124e24469ae3166c3110c62f8',\n",
-       " 'c8dcf956a4f444c6bfe69a00cbeb85ce',\n",
-       " 'd2aeb24c51f549c5b21851be05cb4048',\n",
-       " '6d4c672bafef4db68ae72c8986ee8a72',\n",
-       " 'ff198613768a4361a6d8230b40ee7f78',\n",
-       " 'a133ad72876b48d48e716533c2d78cf0',\n",
-       " 'bdf2b016136e4f57b7ac4b65534031b8',\n",
-       " '5f4f279f304c47839e1a74b0f28de136']"
+       "['4fa8b10c78da4b18ba0830ba8a57367a',\n",
+       " '2eae04b515ee4e9185a9a0e6be812bba',\n",
+       " 'c6039f88486f47f1835ae3b069c5823c',\n",
+       " 'c2c8c51e305144d1917b373125fb4d95',\n",
+       " '79fd23b9ec0648cdab38d1947c6b933e',\n",
+       " '036aa200d8c3492b8a438e4f825f5e7f',\n",
+       " 'c35c77f3ea37460a9a13723fb77b7367',\n",
+       " '6ebccbca571b40d0ab6e83e5e0f2f562',\n",
+       " '38048c2ccc1d4962a4f8f1bd89c8357a',\n",
+       " 'c6b09308360140c7b4f106af3658a31e']"
      ]
     },
-     "execution_count": 17,
+     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
@@ -150,38 +150,58 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 18,
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "[42, 2]"
+      ]
+     },
+     "execution_count": 5,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "# Prepare your documents, metadata, and IDs\n",
+    "docs = [\"Qdrant has Langchain integrations\", \"Qdrant also has Llama Index integrations\"]\n",
+    "metadata = [\n",
+    "    {\"source\": \"Langchain-docs\"},\n",
+    "    {\"source\": \"Linkedin-docs\"},\n",
+    "]\n",
+    "ids = [42, 2]\n",
+    "\n",
+    "# Use the new add method\n",
+    "client.add(collection_name=\"demo_collection\", documents=docs, metadata=metadata, ids=ids)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Behind the scenes, Qdrant Client uses the FastEmbed library to make a passage embedding and then uses the Qdrant API to upsert the documents with metadata, put together as a Points into the collection."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
-      "Top 5 results:\n",
-      "Rank 1: Maharana Pratap was a Rajput warrior king from Mewar. Score: 0.77\n",
-      "Rank 2: Maharana Pratap is considered a symbol of Rajput resistance against foreign rule. Score: 0.77\n",
-      "Rank 3: His legacy is celebrated in Rajasthan through festivals and monuments. Score: 0.69\n",
-      "Rank 4: He had 11 wives and 17 sons, including Amar Singh I who succeeded him as ruler of Mewar. Score: 0.68\n",
-      "Rank 5: He fought against the Mughal Empire led by Akbar. Score: 0.67\n"
+      "[QueryResponse(id=42, embedding=None, metadata={'document': 'Qdrant has Langchain integrations', 'source': 'Langchain-docs'}, document='Qdrant has Langchain integrations', score=0.8276550115796268), QueryResponse(id=2, embedding=None, metadata={'document': 'Qdrant also has Llama Index integrations', 'source': 'Linkedin-docs'}, document='Qdrant also has Llama Index integrations', score=0.8265536935180283)]\n"
     ]
    }
   ],
   "source": [
-    "from qdrant_client.qdrant_fastembed import QueryResponse\n",
-    "\n",
-    "\n",
-    "def print_top_k_results(results: List[QueryResponse], k: int = 5):\n",
-    "    print(f\"Top {k} results:\")\n",
-    "    for i, result in enumerate(results[:k]):\n",
-    "        print(f\"Rank {i + 1}: {result.document}. Score: {result.score:.2f}\")\n",
-    "\n",
-    "\n",
-    "query_text = \"Who is Maharana Pratap?\"\n",
-    "results = client.query(\n",
-    "    collection_name=\"test_collection\", query_text=query_text, limit=7\n",
-    ")  # Returns limit most relevant documents\n",
-    "\n",
-    "print_top_k_results(results)"
+    "search_result = client.query(\n",
+    "    collection_name=\"demo_collection\", query_text=\"This is a query document\"\n",
+    ")\n",
+    "print(search_result)"
   ]
  },
  {
@@ -214,7 +234,7 @@
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
-   "version": "3.9.17"
+   "version": "3.11.5"
  },
  "orig_nbformat": 4
 },
@@ -19,7 +19,7 @@
   "outputs": [],
   "source": [
    "from pathlib import Path\n",
-    "from typing import List, Tuple, Any\n",
+    "from typing import Any\n",
    "\n",
    "import numpy as np\n",
    "import time\n",
@@ -27,7 +27,6 @@
    "from transformers import AutoTokenizer, AutoModel\n",
    "\n",
    "from optimum.onnxruntime import AutoOptimizationConfig, ORTModelForFeatureExtraction, ORTOptimizer\n",
-    "from optimum.onnxruntime.configuration import OptimizationConfig\n",
    "from optimum.pipelines import pipeline\n",
    "import torch.nn.functional as F"
   ]
@@ -92,9 +91,11 @@
    "    return last_hidden.sum(dim=1) / attention_mask.sum(dim=1)[..., None]\n",
    "\n",
    "\n",
-    "def hf_embed(model_id: str, inputs: List[str]):\n",
+    "def hf_embed(model_id: str, inputs: list[str]):\n",
    "    # Tokenize the input texts\n",
-    "    batch_dict = hf_tokenizer(inputs, max_length=512, padding=True, truncation=True, return_tensors=\"pt\")\n",
+    "    batch_dict = hf_tokenizer(\n",
+    "        inputs, max_length=512, padding=True, truncation=True, return_tensors=\"pt\"\n",
+    "    )\n",
    "\n",
    "    outputs = hf_model(**batch_dict)\n",
    "    embeddings = average_pool(outputs.last_hidden_state, batch_dict[\"attention_mask\"])\n",
@@ -134,7 +135,9 @@
    "optimization_config = AutoOptimizationConfig.O4()\n",
    "optimizer = ORTOptimizer.from_pretrained(model)\n",
    "\n",
-    "optimizer.optimize(save_dir=save_dir, optimization_config=optimization_config, use_external_data_format=True)\n",
+    "optimizer.optimize(\n",
+    "    save_dir=save_dir, optimization_config=optimization_config, use_external_data_format=True\n",
+    ")\n",
    "model = ORTModelForFeatureExtraction.from_pretrained(save_dir)\n",
    "\n",
    "tokenizer.save_pretrained(save_dir)\n",
@@ -149,7 +152,9 @@
   "metadata": {},
   "outputs": [],
   "source": [
-    "onnx_quant_embed = pipeline(\"feature-extraction\", model=model, accelerator=\"ort\", tokenizer=tokenizer,return_tensors=True)"
+    "onnx_quant_embed = pipeline(\n",
+    "    \"feature-extraction\", model=model, accelerator=\"ort\", tokenizer=tokenizer, return_tensors=True\n",
+    ")"
   ]
  },
  {
@@ -159,9 +164,8 @@
   "metadata": {},
   "outputs": [],
   "source": [
-    "import torch\n",
    "embeddings = onnx_quant_embed(inputs=english_texts)\n",
-    "F.normalize(embeddings[4])[:,0], english_texts[4], len(embeddings), len(english_texts)"
+    "F.normalize(embeddings[4])[:, 0], english_texts[4], len(embeddings), len(english_texts)"
   ]
  },
  {
@@ -171,8 +175,9 @@
   "metadata": {},
   "outputs": [],
   "source": [
-    "\n",
-    "def measure_pipeline_time(pipeline, input_texts: List[str], num_runs=10, **kwargs: Any) -> Tuple[float, float]:\n",
+    "def measure_pipeline_time(\n",
+    "    pipeline, input_texts: list[str], num_runs=10, **kwargs: Any\n",
+    ") -> tuple[float, float]:\n",
    "    \"\"\"Measures the time it takes to run the pipeline on the input texts.\"\"\"\n",
    "    times = []\n",
    "    total_chars = sum(len(text) for text in input_texts)\n",
@@ -256,6 +261,7 @@
    "\n",
    "save_dir = Path(\"../local_cache/fast-bge-small-en-v1.5\")\n",
    "\n",
+    "\n",
    "def compress(directory_path):\n",
    "    directory_path = Path(directory_path)\n",
    "    assert directory_path.exists(), f\"{directory_path} does not exist\"\n",
@@ -304,9 +310,9 @@
    }
   ],
   "source": [
-    "import os\n",
    "from google.cloud import storage\n",
    "\n",
+    "\n",
    "def upload(bucket_name, source_file_path):\n",
    "    storage_client = storage.Client(project=\"main\")\n",
    "    bucket = storage_client.bucket(bucket_name)\n",
@@ -0,0 +1,371 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import numpy as np\n",
+    "import torch\n",
+    "from transformers import AutoModelForMaskedLM, AutoTokenizer"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Running the model with Transformers and Torch"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "sentences = [\n",
+    "    \"Hello World\",\n",
+    "    \"Built by Nirant Kasliwal\",\n",
+    "]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## PyTorch Code from the [SPLADERunner](https://github.com/PrithivirajDamodaran/SPLADERunner) library"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "hf_token = \"<your_hf_token_here>\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Output Logits shape:  torch.Size([2, 10, 30522])\n",
+      "Output Attention mask shape:  torch.Size([2, 10])\n",
+      "Sparse Vector shape:  torch.Size([2, 30522])\n",
+      "SPLADE BOW rep for sentence:\tBuilt by Nirant Kasliwal\n",
+      "[('##rant', 2.02), ('built', 1.94), ('##wal', 1.79), ('##sl', 1.69), ('build', 1.57), ('ka', 1.4), ('ni', 1.26), ('made', 0.93), ('architect', 0.76), ('was', 0.69), ('who', 0.61), ('his', 0.5), ('wrote', 0.47), ('india', 0.45), ('company', 0.41), ('##i', 0.41), ('he', 0.37), ('manufacturer', 0.36), ('by', 0.35), ('engineer', 0.33), ('architecture', 0.33), ('ko', 0.23), ('him', 0.22), ('invented', 0.19), ('said', 0.14), ('k', 0.11), ('man', 0.11), ('statue', 0.11), ('bomb', 0.1), ('##wa', 0.1), ('builder', 0.09), ('.', 0.07), ('started', 0.06), (',', 0.04), ('ku', 0.03)]\n"
+     ]
+    }
+   ],
+   "source": [
+    "# Download the model and tokenizer\n",
+    "device = \"cuda:0\" if torch.cuda.is_available() else \"cpu\"\n",
+    "tokenizer = AutoTokenizer.from_pretrained(\"prithivida/Splade_PP_en_v1\", token=hf_token)\n",
+    "reverse_voc = {v: k for k, v in tokenizer.vocab.items()}\n",
+    "model = AutoModelForMaskedLM.from_pretrained(\"prithivida/Splade_PP_en_v1\", token=hf_token)\n",
+    "model.to(device)\n",
+    "\n",
+    "# Tokenize the input\n",
+    "inputs = tokenizer(sentences, return_tensors=\"pt\", padding=True, truncation=True, max_length=512)\n",
+    "inputs = {key: val.to(device) for key, val in inputs.items()}\n",
+    "input_ids = inputs[\"input_ids\"]\n",
+    "attention_mask = inputs[\"attention_mask\"]\n",
+    "token_type_ids = inputs[\"token_type_ids\"]\n",
+    "\n",
+    "# Run model and prepare sparse vector\n",
+    "outputs = model(**inputs)\n",
+    "logits = outputs.logits\n",
+    "print(\"Output Logits shape: \", logits.shape)\n",
+    "print(\"Output Attention mask shape: \", attention_mask.shape)\n",
+    "relu_log = torch.log(1 + torch.relu(logits))\n",
+    "weighted_log = relu_log * attention_mask.unsqueeze(-1)\n",
+    "max_val, _ = torch.max(weighted_log, dim=1)\n",
+    "vector = max_val.squeeze()\n",
+    "print(\"Sparse Vector shape: \", vector.shape)\n",
+    "# print(\"Number of Actual Dimensions: \", len(cols))\n",
+    "cols = [vec.nonzero().squeeze().cpu().tolist() for vec in vector]\n",
+    "weights = [vec[col].cpu().tolist() for vec, col in zip(vector, cols)]\n",
+    "\n",
+    "idx = 1\n",
+    "cols, weights = cols[idx], weights[idx]\n",
+    "# Print the BOW representation\n",
+    "d = {k: v for k, v in zip(cols, weights)}\n",
+    "sorted_d = {k: v for k, v in sorted(d.items(), key=lambda item: item[1], reverse=True)}\n",
+    "bow_rep = []\n",
+    "for k, v in sorted_d.items():\n",
+    "    bow_rep.append((reverse_voc[k], round(v, 2)))\n",
+    "print(f\"SPLADE BOW rep for sentence:\\t{sentences[idx]}\\n{bow_rep}\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Export with output_attentions and logits"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Exporting model to models/nirantk_SPLADE_PP_en_v1\n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "('models/nirantk_SPLADE_PP_en_v1/tokenizer_config.json',\n",
+       " 'models/nirantk_SPLADE_PP_en_v1/special_tokens_map.json',\n",
+       " 'models/nirantk_SPLADE_PP_en_v1/vocab.txt',\n",
+       " 'models/nirantk_SPLADE_PP_en_v1/added_tokens.json',\n",
+       " 'models/nirantk_SPLADE_PP_en_v1/tokenizer.json')"
+      ]
+     },
+     "execution_count": 5,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "from transformers import AutoTokenizer\n",
+    "\n",
+    "model_id = \"nirantk/SPLADE_PP_en_v1\"\n",
+    "output_dir = f\"models/{model_id.replace('/', '_')}\"\n",
+    "model_kwargs = {\"output_attentions\": True, \"return_dict\": True}\n",
+    "\n",
+    "print(f\"Exporting model to {output_dir}\")\n",
+    "tokenizer.save_pretrained(output_dir)\n",
+    "# main_export(\n",
+    "#     model_id,\n",
+    "#     output=output_dir,\n",
+    "#     no_post_process=True,\n",
+    "#     model_kwargs=model_kwargs,\n",
+    "#     token=hf_token,\n",
+    "# )"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Running the model with ONNX"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from optimum.onnxruntime import ORTModelForMaskedLM\n",
+    "\n",
+    "model = ORTModelForMaskedLM.from_pretrained(\"nirantk/SPLADE_PP_en_v1\")\n",
+    "tokenizer = AutoTokenizer.from_pretrained(\"nirantk/SPLADE_PP_en_v1\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "inputs = tokenizer(sentences, return_tensors=\"pt\", padding=True, truncation=True, max_length=512)\n",
+    "inputs = {key: val.to(device) for key, val in inputs.items()}\n",
+    "input_ids = inputs[\"input_ids\"]\n",
+    "attention_mask = inputs[\"attention_mask\"]\n",
+    "token_type_ids = inputs[\"token_type_ids\"]\n",
+    "\n",
+    "onnx_input = {\n",
+    "    \"input_ids\": input_ids.cpu().numpy(),\n",
+    "    \"attention_mask\": attention_mask.cpu().numpy(),\n",
+    "    \"token_type_ids\": token_type_ids.cpu().numpy(),\n",
+    "}\n",
+    "\n",
+    "logits = model(**onnx_input).logits"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "(2, 10, 30522)"
+      ]
+     },
+     "execution_count": 8,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "logits.shape"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Output Logits shape:  (2, 10, 30522)\n",
+      "Sparse Vector shape:  (2, 30522)\n",
+      "SPLADE BOW rep for sentence:\tBuilt by Nirant Kasliwal\n",
+      "[('##rant', 2.02), ('built', 1.94), ('##wal', 1.79), ('##sl', 1.69), ('build', 1.57), ('ka', 1.4), ('ni', 1.26), ('made', 0.93), ('architect', 0.76), ('was', 0.69), ('who', 0.61), ('his', 0.5), ('wrote', 0.47), ('india', 0.45), ('company', 0.41), ('##i', 0.41), ('he', 0.37), ('manufacturer', 0.36), ('by', 0.35), ('engineer', 0.33), ('architecture', 0.33), ('ko', 0.23), ('him', 0.22), ('invented', 0.19), ('said', 0.14), ('k', 0.11), ('man', 0.11), ('statue', 0.11), ('bomb', 0.1), ('##wa', 0.1), ('builder', 0.09), ('.', 0.07), ('started', 0.06), (',', 0.04), ('ku', 0.03)]\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(\"Output Logits shape: \", logits.shape)\n",
+    "\n",
+    "relu_log = np.log(1 + np.maximum(logits, 0))\n",
+    "\n",
+    "# Equivalent to relu_log * attention_mask.unsqueeze(-1)\n",
+    "# For NumPy, you might need to explicitly expand dimensions if 'attention_mask' is not already 2D\n",
+    "weighted_log = relu_log * np.expand_dims(attention_mask, axis=-1)\n",
+    "\n",
+    "# Equivalent to torch.max(weighted_log, dim=1)\n",
+    "# NumPy's max function returns only the max values, not the indices, so we don't need to unpack two values\n",
+    "max_val = np.max(weighted_log, axis=1)\n",
+    "\n",
+    "# Equivalent to max_val.squeeze()\n",
+    "# This step may be unnecessary in NumPy if max_val doesn't have unnecessary dimensions\n",
+    "vector = np.squeeze(max_val)\n",
+    "print(\"Sparse Vector shape: \", vector.shape)\n",
+    "\n",
+    "# print(vector[0].nonzero())\n",
+    "\n",
+    "cols = [vec.nonzero()[0].squeeze().tolist() for vec in vector]\n",
+    "weights = [vec[col].tolist() for vec, col in zip(vector, cols)]\n",
+    "\n",
+    "idx = 1\n",
+    "cols, weights = cols[idx], weights[idx]\n",
+    "# Print the BOW representation\n",
+    "d = {k: v for k, v in zip(cols, weights)}\n",
+    "sorted_d = {k: v for k, v in sorted(d.items(), key=lambda item: item[1], reverse=True)}\n",
+    "bow_rep = []\n",
+    "for k, v in sorted_d.items():\n",
+    "    bow_rep.append((reverse_voc[k], round(v, 2)))\n",
+    "print(f\"SPLADE BOW rep for sentence:\\t{sentences[idx]}\\n{bow_rep}\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 10,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "35"
+      ]
+     },
+     "execution_count": 10,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "len(cols)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 11,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "[1010,\n",
+       " 1012,\n",
+       " 1047,\n",
+       " 2001,\n",
+       " 2002,\n",
+       " 2010,\n",
+       " 2011,\n",
+       " 2032,\n",
+       " 2040,\n",
+       " 2056,\n",
+       " 2072,\n",
+       " 2081,\n",
+       " 2158,\n",
+       " 2194,\n",
+       " 2318,\n",
+       " 2328,\n",
+       " 2626,\n",
+       " 2634,\n",
+       " 3857,\n",
+       " 3992,\n",
+       " 4213,\n",
+       " 4294,\n",
+       " 4944,\n",
+       " 5968,\n",
+       " 6231,\n",
+       " 7751,\n",
+       " 8826,\n",
+       " 9152,\n",
+       " 10556,\n",
+       " 12508,\n",
+       " 12849,\n",
+       " 13476,\n",
+       " 13970,\n",
+       " 14540,\n",
+       " 17884]"
+      ]
+     },
+     "execution_count": 11,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "cols"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "fst",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.13"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}
@@ -0,0 +1,122 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "4bdb2a91-fa2a-4cee-ad5a-176cc957394d",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-05-23T12:15:28.171586Z",
+     "start_time": "2024-05-23T12:15:28.076314Z"
+    }
+   },
+   "outputs": [
+    {
+     "ename": "ModuleNotFoundError",
+     "evalue": "No module named 'torch'",
+     "output_type": "error",
+     "traceback": [
+      "\u001B[0;31m---------------------------------------------------------------------------\u001B[0m",
+      "\u001B[0;31mModuleNotFoundError\u001B[0m                       Traceback (most recent call last)",
+      "Cell \u001B[0;32mIn[1], line 1\u001B[0m\n\u001B[0;32m----> 1\u001B[0m \u001B[38;5;28;01mimport\u001B[39;00m \u001B[38;5;21;01mtorch\u001B[39;00m\n\u001B[1;32m      2\u001B[0m \u001B[38;5;28;01mimport\u001B[39;00m \u001B[38;5;21;01mtorch\u001B[39;00m\u001B[38;5;21;01m.\u001B[39;00m\u001B[38;5;21;01monnx\u001B[39;00m\n\u001B[1;32m      3\u001B[0m \u001B[38;5;28;01mimport\u001B[39;00m \u001B[38;5;21;01mtorchvision\u001B[39;00m\u001B[38;5;21;01m.\u001B[39;00m\u001B[38;5;21;01mmodels\u001B[39;00m \u001B[38;5;28;01mas\u001B[39;00m \u001B[38;5;21;01mmodels\u001B[39;00m\n",
+      "\u001B[0;31mModuleNotFoundError\u001B[0m: No module named 'torch'"
+     ]
+    }
+   ],
+   "source": [
+    "import torch\n",
+    "import torch.onnx\n",
+    "import torchvision.models as models\n",
+    "import torchvision.transforms as transforms\n",
+    "from PIL import Image\n",
+    "import numpy as np\n",
+    "from tests.config import TEST_MISC_DIR\n",
+    "\n",
+    "# Load pre-trained ResNet-50 model\n",
+    "resnet = models.resnet50(pretrained=True)\n",
+    "resnet = torch.nn.Sequential(*(list(resnet.children())[:-1]))  # Remove the last fully connected layer\n",
+    "resnet.eval()\n",
+    "\n",
+    "# Define preprocessing transform\n",
+    "preprocess = transforms.Compose([\n",
+    "    transforms.Resize(256),\n",
+    "    transforms.CenterCrop(224),\n",
+    "    transforms.ToTensor(),\n",
+    "    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),\n",
+    "])\n",
+    "\n",
+    "# Load and preprocess the image\n",
+    "def preprocess_image(image_path):\n",
+    "    input_image = Image.open(image_path)\n",
+    "    input_tensor = preprocess(input_image)\n",
+    "    input_batch = input_tensor.unsqueeze(0)  # Add batch dimension\n",
+    "    return input_batch\n",
+    "\n",
+    "# Example input for exporting\n",
+    "input_image = preprocess_image('example.jpg')\n",
+    "\n",
+    "# Export the model to ONNX with dynamic axes\n",
+    "torch.onnx.export(\n",
+    "    resnet, \n",
+    "    input_image, \n",
+    "    \"model.onnx\", \n",
+    "    export_params=True, \n",
+    "    opset_version=9, \n",
+    "    input_names=['input'], \n",
+    "    output_names=['output'],\n",
+    "    dynamic_axes={'input': {0: 'batch_size'}, 'output': {0: 'batch_size'}}\n",
+    ")\n",
+    "\n",
+    "# Load ONNX model\n",
+    "import onnx\n",
+    "import onnxruntime as ort\n",
+    "\n",
+    "onnx_model = onnx.load(\"model.onnx\")\n",
+    "ort_session = ort.InferenceSession(\"model.onnx\")\n",
+    "\n",
+    "# Run inference and extract feature vectors\n",
+    "def extract_feature_vectors(image_paths):\n",
+    "    input_images = [preprocess_image(image_path) for image_path in image_paths]\n",
+    "    input_batch = torch.cat(input_images, dim=0)  # Combine images into a single batch\n",
+    "    ort_inputs = {ort_session.get_inputs()[0].name: input_batch.numpy()}\n",
+    "    ort_outs = ort_session.run(None, ort_inputs)\n",
+    "    return ort_outs[0]\n",
+    "\n",
+    "# Example usage\n",
+    "images = [TEST_MISC_DIR / \"image.jpeg\", str(TEST_MISC_DIR / \"small_image.jpeg\")]  # Replace with your image paths\n",
+    "feature_vectors = extract_feature_vectors(images)\n",
+    "print(\"Feature vector shape:\", feature_vectors.shape)\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "outputs": [],
+   "source": [],
+   "metadata": {
+    "collapsed": false
+   },
+   "id": "baa650c4cb3e0e6d"
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.2"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}
@@ -0,0 +1,17 @@
+from optimum.exporters.onnx import main_export
+from transformers import AutoTokenizer
+
+model_id = "sentence-transformers/paraphrase-MiniLM-L6-v2"
+output_dir = f"models/{model_id.replace('/', '_')}"
+model_kwargs = {"output_attentions": True, "return_dict": True}
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+
+# export if the output model does not exist
+# try:
+#     sess = onnxruntime.InferenceSession(f"{output_dir}/model.onnx")
+#     print("Model already exported")
+# except FileNotFoundError:
+print(f"Exporting model to {output_dir}")
+main_export(
+    model_id, output=output_dir, no_post_process=True, model_kwargs=model_kwargs
+)
@@ -0,0 +1,33 @@
+import numpy as np
+import onnx
+import onnxruntime
+from transformers import AutoTokenizer
+
+model_id = "sentence-transformers/paraphrase-MiniLM-L6-v2"
+output_dir = f"models/{model_id.replace('/', '_')}"
+model_kwargs = {"output_attentions": True, "return_dict": True}
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+
+model_path = f"{output_dir}/model.onnx"
+onnx_model = onnx.load(model_path)
+ort_session = onnxruntime.InferenceSession(model_path)
+text = "This is a test sentence"
+tokenizer_output = tokenizer(text, return_tensors="np")
+input_ids = tokenizer_output["input_ids"]
+attention_mask = tokenizer_output["attention_mask"]
+print(attention_mask)
+# Prepare the input
+input_ids = np.array(input_ids).astype(
+    np.int64
+)  # Replace your_input_ids with actual input data
+
+# Run the ONNX model
+outputs = ort_session.run(
+    None, {"input_ids": input_ids, "attention_mask": attention_mask}
+)
+
+# Get the attention weights
+attentions = outputs[-1]
+
+# Print the attention weights for the first layer and first head
+print(attentions[0][0])
@@ -0,0 +1,22 @@
+import importlib.metadata
+
+from fastembed.image import ImageEmbedding
+from fastembed.late_interaction import LateInteractionTextEmbedding
+from fastembed.late_interaction_multimodal import LateInteractionMultimodalEmbedding
+from fastembed.sparse import SparseEmbedding, SparseTextEmbedding
+from fastembed.text import TextEmbedding
+
+try:
+    version = importlib.metadata.version("fastembed")
+except importlib.metadata.PackageNotFoundError as _:
+    version = importlib.metadata.version("fastembed-gpu")
+
+__version__ = version
+__all__ = [
+    "TextEmbedding",
+    "SparseTextEmbedding",
+    "SparseEmbedding",
+    "ImageEmbedding",
+    "LateInteractionTextEmbedding",
+    "LateInteractionMultimodalEmbedding",
+]
@@ -0,0 +1,3 @@
+from fastembed.common.types import ImageInput, OnnxProvider, PathInput
+
+__all__ = ["OnnxProvider", "ImageInput", "PathInput"]
@@ -0,0 +1,52 @@
+from dataclasses import dataclass, field
+from enum import Enum
+from typing import Optional, Any
+
+
+@dataclass(frozen=True)
+class ModelSource:
+    hf: Optional[str] = None
+    url: Optional[str] = None
+    _deprecated_tar_struct: bool = False
+
+    @property
+    def deprecated_tar_struct(self) -> bool:
+        return self._deprecated_tar_struct
+
+    def __post_init__(self) -> None:
+        if self.hf is None and self.url is None:
+            raise ValueError(
+                f"At least one source should be set, current sources: hf={self.hf}, url={self.url}"
+            )
+
+
+@dataclass(frozen=True)
+class BaseModelDescription:
+    model: str
+    sources: ModelSource
+    model_file: str
+    description: str
+    license: str
+    size_in_GB: float
+    additional_files: list[str] = field(default_factory=list)
+
+
+@dataclass(frozen=True)
+class DenseModelDescription(BaseModelDescription):
+    dim: Optional[int] = None
+    tasks: Optional[dict[str, Any]] = field(default_factory=dict)
+
+    def __post_init__(self) -> None:
+        assert self.dim is not None, "dim is required for dense model description"
+
+
+@dataclass(frozen=True)
+class SparseModelDescription(BaseModelDescription):
+    requires_idf: Optional[bool] = None
+    vocab_size: Optional[int] = None
+
+
+class PoolingType(str, Enum):
+    CLS = "CLS"
+    MEAN = "MEAN"
+    DISABLED = "DISABLED"
@@ -0,0 +1,458 @@
+import os
+import time
+import json
+import shutil
+import tarfile
+from pathlib import Path
+from typing import Any, Optional, Union, TypeVar, Generic
+
+import requests
+from huggingface_hub import snapshot_download, model_info, list_repo_tree
+from huggingface_hub.hf_api import RepoFile
+from huggingface_hub.utils import (
+    RepositoryNotFoundError,
+    disable_progress_bars,
+    enable_progress_bars,
+)
+from loguru import logger
+from tqdm import tqdm
+from fastembed.common.model_description import BaseModelDescription
+
+T = TypeVar("T", bound=BaseModelDescription)
+
+
+class ModelManagement(Generic[T]):
+    METADATA_FILE = "files_metadata.json"
+
+    @classmethod
+    def list_supported_models(cls) -> list[dict[str, Any]]:
+        """Lists the supported models.
+
+        Returns:
+            list[T]: A list of dictionaries containing the model information.
+        """
+        raise NotImplementedError()
+
+    @classmethod
+    def add_custom_model(
+        cls,
+        *args: Any,
+        **kwargs: Any,
+    ) -> None:
+        """Add a custom model to the existing embedding classes based on the passed model descriptions
+
+        Model description dict should contain the fields same as in one of the model descriptions presented
+         in fastembed.common.model_description
+
+         E.g. for BaseModelDescription:
+              model: str
+              sources: ModelSource
+              model_file: str
+              description: str
+              license: str
+              size_in_GB: float
+              additional_files: list[str]
+
+        Returns:
+            None
+        """
+        raise NotImplementedError()
+
+    @classmethod
+    def _list_supported_models(cls) -> list[T]:
+        raise NotImplementedError()
+
+    @classmethod
+    def _get_model_description(cls, model_name: str) -> T:
+        """
+        Gets the model description from the model_name.
+
+        Args:
+            model_name (str): The name of the model.
+
+        raises:
+            ValueError: If the model_name is not supported.
+
+        Returns:
+            T: The model description.
+        """
+        for model in cls._list_supported_models():
+            if model_name.lower() == model.model.lower():
+                return model
+
+        raise ValueError(f"Model {model_name} is not supported in {cls.__name__}.")
+
+    @classmethod
+    def download_file_from_gcs(cls, url: str, output_path: str, show_progress: bool = True) -> str:
+        """
+        Downloads a file from Google Cloud Storage.
+
+        Args:
+            url (str): The URL to download the file from.
+            output_path (str): The path to save the downloaded file to.
+            show_progress (bool, optional): Whether to show a progress bar. Defaults to True.
+
+        Returns:
+            str: The path to the downloaded file.
+        """
+
+        if os.path.exists(output_path):
+            return output_path
+        response = requests.get(url, stream=True)
+
+        # Handle HTTP errors
+        if response.status_code == 403:
+            raise PermissionError(
+                "Authentication Error: You do not have permission to access this resource. "
+                "Please check your credentials."
+            )
+
+        # Get the total size of the file
+        total_size_in_bytes = int(response.headers.get("content-length", 0))
+
+        # Warn if the total size is zero
+        if total_size_in_bytes == 0:
+            print(f"Warning: Content-length header is missing or zero in the response from {url}.")
+
+        show_progress = bool(total_size_in_bytes and show_progress)
+
+        with tqdm(
+            total=total_size_in_bytes,
+            unit="iB",
+            unit_scale=True,
+            disable=not show_progress,
+        ) as progress_bar:
+            with open(output_path, "wb") as file:
+                for chunk in response.iter_content(chunk_size=1024):
+                    if chunk:  # Filter out keep-alive new chunks
+                        progress_bar.update(len(chunk))
+                        file.write(chunk)
+        return output_path
+
+    @classmethod
+    def download_files_from_huggingface(
+        cls,
+        hf_source_repo: str,
+        cache_dir: str,
+        extra_patterns: list[str],
+        local_files_only: bool = False,
+        **kwargs: Any,
+    ) -> str:
+        """
+        Downloads a model from HuggingFace Hub.
+        Args:
+            hf_source_repo (str): Name of the model on HuggingFace Hub, e.g. "qdrant/all-MiniLM-L6-v2-onnx".
+            cache_dir (Optional[str]): The path to the cache directory.
+            extra_patterns (list[str]): extra patterns to allow in the snapshot download, typically
+                includes the required model files.
+            local_files_only (bool, optional): Whether to only use local files. Defaults to False.
+        Returns:
+            Path: The path to the model directory.
+        """
+
+        def _verify_files_from_metadata(
+            model_dir: Path, stored_metadata: dict[str, Any], repo_files: list[RepoFile]
+        ) -> bool:
+            try:
+                for rel_path, meta in stored_metadata.items():
+                    file_path = model_dir / rel_path
+
+                    if not file_path.exists():
+                        return False
+
+                    if repo_files:  # online verification
+                        file_info = next((f for f in repo_files if f.path == file_path.name), None)
+                        if (
+                            not file_info
+                            or file_info.size != meta["size"]
+                            or file_info.blob_id != meta["blob_id"]
+                        ):
+                            return False
+
+                    else:  # offline verification
+                        if file_path.stat().st_size != meta["size"]:
+                            return False
+                return True
+            except (OSError, KeyError) as e:
+                logger.error(f"Error verifying files: {str(e)}")
+                return False
+
+        def _collect_file_metadata(
+            model_dir: Path, repo_files: list[RepoFile]
+        ) -> dict[str, dict[str, Union[int, str]]]:
+            meta: dict[str, dict[str, Union[int, str]]] = {}
+            file_info_map = {f.path: f for f in repo_files}
+            for file_path in model_dir.rglob("*"):
+                if file_path.is_file() and file_path.name != cls.METADATA_FILE:
+                    repo_file = file_info_map.get(file_path.name)
+                    if repo_file:
+                        meta[str(file_path.relative_to(model_dir))] = {
+                            "size": repo_file.size,
+                            "blob_id": repo_file.blob_id,
+                        }
+            return meta
+
+        def _save_file_metadata(
+            model_dir: Path, meta: dict[str, dict[str, Union[int, str]]]
+        ) -> None:
+            try:
+                if not model_dir.exists():
+                    model_dir.mkdir(parents=True, exist_ok=True)
+                (model_dir / cls.METADATA_FILE).write_text(json.dumps(meta))
+            except (OSError, ValueError) as e:
+                logger.warning(f"Error saving metadata: {str(e)}")
+
+        allow_patterns = [
+            "config.json",
+            "tokenizer.json",
+            "tokenizer_config.json",
+            "special_tokens_map.json",
+            "preprocessor_config.json",
+        ]
+
+        allow_patterns.extend(extra_patterns)
+
+        snapshot_dir = Path(cache_dir) / f"models--{hf_source_repo.replace('/', '--')}"
+        metadata_file = snapshot_dir / cls.METADATA_FILE
+
+        if local_files_only:
+            disable_progress_bars()
+            if metadata_file.exists():
+                metadata = json.loads(metadata_file.read_text())
+                verified = _verify_files_from_metadata(snapshot_dir, metadata, repo_files=[])
+                if not verified:
+                    logger.warning(
+                        "Local file sizes do not match the metadata."
+                    )  # do not raise, still make an attempt to load the model
+            else:
+                logger.warning(
+                    "Metadata file not found. Proceeding without checking local files."
+                )  # if users have downloaded models from hf manually, or they're updating from previous versions of
+                # fastembed
+            result = snapshot_download(
+                repo_id=hf_source_repo,
+                allow_patterns=allow_patterns,
+                cache_dir=cache_dir,
+                local_files_only=local_files_only,
+                **kwargs,
+            )
+            return result
+
+        repo_revision = model_info(hf_source_repo).sha
+        repo_tree = list(list_repo_tree(hf_source_repo, revision=repo_revision, repo_type="model"))
+
+        allowed_extensions = {".json", ".onnx", ".txt"}
+        repo_files = (
+            [
+                f
+                for f in repo_tree
+                if isinstance(f, RepoFile) and Path(f.path).suffix in allowed_extensions
+            ]
+            if repo_tree
+            else []
+        )
+
+        verified_metadata = False
+
+        if snapshot_dir.exists() and metadata_file.exists():
+            metadata = json.loads(metadata_file.read_text())
+            verified_metadata = _verify_files_from_metadata(snapshot_dir, metadata, repo_files)
+
+        if verified_metadata:
+            disable_progress_bars()
+
+        result = snapshot_download(
+            repo_id=hf_source_repo,
+            allow_patterns=allow_patterns,
+            cache_dir=cache_dir,
+            local_files_only=local_files_only,
+            **kwargs,
+        )
+
+        if (
+            not verified_metadata
+        ):  # metadata is not up-to-date, update it and check whether the files have been
+            # downloaded correctly
+            metadata = _collect_file_metadata(snapshot_dir, repo_files)
+
+            download_successful = _verify_files_from_metadata(
+                snapshot_dir, metadata, repo_files=[]
+            )  # offline verification
+            if not download_successful:
+                raise ValueError(
+                    "Files have been corrupted during downloading process. "
+                    "Please check your internet connection and try again."
+                )
+            _save_file_metadata(snapshot_dir, metadata)
+
+        return result
+
+    @classmethod
+    def decompress_to_cache(cls, targz_path: str, cache_dir: str) -> str:
+        """
+        Decompresses a .tar.gz file to a cache directory.
+
+        Args:
+            targz_path (str): Path to the .tar.gz file.
+            cache_dir (str): Path to the cache directory.
+
+        Returns:
+            cache_dir (str): Path to the cache directory.
+        """
+        # Check if targz_path exists and is a file
+        if not os.path.isfile(targz_path):
+            raise ValueError(f"{targz_path} does not exist or is not a file.")
+
+        # Check if targz_path is a .tar.gz file
+        if not targz_path.endswith(".tar.gz"):
+            raise ValueError(f"{targz_path} is not a .tar.gz file.")
+
+        try:
+            # Open the tar.gz file
+            with tarfile.open(targz_path, "r:gz") as tar:
+                # Extract all files into the cache directory
+                tar.extractall(
+                    path=cache_dir,
+                )
+        except tarfile.TarError as e:
+            # If any error occurs while opening or extracting the tar.gz file,
+            # delete the cache directory (if it was created in this function)
+            # and raise the error again
+            if "tmp" in cache_dir:
+                shutil.rmtree(cache_dir)
+            raise ValueError(f"An error occurred while decompressing {targz_path}: {e}")
+
+        return cache_dir
+
+    @classmethod
+    def retrieve_model_gcs(
+        cls,
+        model_name: str,
+        source_url: str,
+        cache_dir: str,
+        deprecated_tar_struct: bool = False,
+        local_files_only: bool = False,
+    ) -> Path:
+        fast_model_name = f"{'fast-' if deprecated_tar_struct else ''}{model_name.split('/')[-1]}"
+        cache_tmp_dir = Path(cache_dir) / "tmp"
+        model_tmp_dir = cache_tmp_dir / fast_model_name
+        model_dir = Path(cache_dir) / fast_model_name
+
+        # check if the model_dir and the model files are both present for macOS
+        if model_dir.exists() and len(list(model_dir.glob("*"))) > 0:
+            return model_dir
+
+        if model_tmp_dir.exists():
+            shutil.rmtree(model_tmp_dir)
+
+        cache_tmp_dir.mkdir(parents=True, exist_ok=True)
+
+        model_tar_gz = Path(cache_dir) / f"{fast_model_name}.tar.gz"
+
+        if model_tar_gz.exists():
+            model_tar_gz.unlink()
+
+        if not local_files_only:
+            cls.download_file_from_gcs(
+                source_url,
+                output_path=str(model_tar_gz),
+            )
+
+            cls.decompress_to_cache(targz_path=str(model_tar_gz), cache_dir=str(cache_tmp_dir))
+            assert model_tmp_dir.exists(), f"Could not find {model_tmp_dir} in {cache_tmp_dir}"
+
+            model_tar_gz.unlink()
+            # Rename from tmp to final name is atomic
+            model_tmp_dir.rename(model_dir)
+        else:
+            logger.error(
+                f"Could not find the model tar.gz file at {model_dir} and local_files_only=True."
+            )
+            raise ValueError(
+                f"Could not find the model tar.gz file at {model_dir} and local_files_only=True."
+            )
+
+        return model_dir
+
+    @classmethod
+    def download_model(cls, model: T, cache_dir: str, retries: int = 3, **kwargs: Any) -> Path:
+        """
+        Downloads a model from HuggingFace Hub or Google Cloud Storage.
+
+        Args:
+            model (T): The model description.
+                Example:
+                ```
+                {
+                    "model": "BAAI/bge-base-en-v1.5",
+                    "dim": 768,
+                    "description": "Base English model, v1.5",
+                    "size_in_GB": 0.44,
+                    "sources": {
+                        "url": "https://storage.googleapis.com/qdrant-fastembed/fast-bge-base-en-v1.5.tar.gz",
+                        "hf": "qdrant/bge-base-en-v1.5-onnx-q",
+                    }
+                }
+                ```
+            cache_dir (str): The path to the cache directory.
+            retries: (int): The number of times to retry (including the first attempt)
+
+        Returns:
+            Path: The path to the downloaded model directory.
+        """
+        local_files_only = kwargs.get("local_files_only", False)
+        specific_model_path: Optional[str] = kwargs.pop("specific_model_path", None)
+        if specific_model_path:
+            return Path(specific_model_path)
+        retries = 1 if local_files_only else retries
+        hf_source = model.sources.hf
+        url_source = model.sources.url
+
+        sleep = 3.0
+        while retries > 0:
+            retries -= 1
+
+            if hf_source:
+                extra_patterns = [model.model_file]
+                extra_patterns.extend(model.additional_files)
+
+                try:
+                    return Path(
+                        cls.download_files_from_huggingface(
+                            hf_source,
+                            cache_dir=cache_dir,
+                            extra_patterns=extra_patterns,
+                            **kwargs,
+                        )
+                    )
+                except (EnvironmentError, RepositoryNotFoundError, ValueError) as e:
+                    if not local_files_only:
+                        logger.error(
+                            f"Could not download model from HuggingFace: {e} "
+                            "Falling back to other sources."
+                        )
+                finally:
+                    enable_progress_bars()
+            if url_source or local_files_only:
+                try:
+                    return cls.retrieve_model_gcs(
+                        model.model,
+                        str(url_source),
+                        str(cache_dir),
+                        deprecated_tar_struct=model.sources.deprecated_tar_struct,
+                        local_files_only=local_files_only,
+                    )
+                except Exception:
+                    if not local_files_only:
+                        logger.error(f"Could not download model from url: {url_source}")
+
+            if local_files_only:
+                logger.error("Could not find model in cache_dir")
+            else:
+                logger.error(
+                    f"Could not download model from either source, sleeping for {sleep} seconds, {retries} retries left."
+                )
+            time.sleep(sleep)
+            sleep *= 3
+
+        raise ValueError(f"Could not load model {model.model} from any source.")
@@ -0,0 +1,136 @@
+import warnings
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any, Generic, Iterable, Optional, Sequence, Type, TypeVar
+
+import numpy as np
+import onnxruntime as ort
+
+from numpy.typing import NDArray
+from tokenizers import Tokenizer
+
+from fastembed.common.types import OnnxProvider, NumpyArray
+from fastembed.parallel_processor import Worker
+
+# Holds type of the embedding result
+T = TypeVar("T")
+
+
+@dataclass
+class OnnxOutputContext:
+    model_output: NumpyArray
+    attention_mask: Optional[NDArray[np.int64]] = None
+    input_ids: Optional[NDArray[np.int64]] = None
+
+
+class OnnxModel(Generic[T]):
+    @classmethod
+    def _get_worker_class(cls) -> Type["EmbeddingWorker[T]"]:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[T]:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def __init__(self) -> None:
+        self.model: Optional[ort.InferenceSession] = None
+        self.tokenizer: Optional[Tokenizer] = None
+
+    def _preprocess_onnx_input(
+        self, onnx_input: dict[str, NumpyArray], **kwargs: Any
+    ) -> dict[str, NumpyArray]:
+        """
+        Preprocess the onnx input.
+        """
+        return onnx_input
+
+    def _load_onnx_model(
+        self,
+        model_dir: Path,
+        model_file: str,
+        threads: Optional[int],
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_id: Optional[int] = None,
+    ) -> None:
+        model_path = model_dir / model_file
+        # List of Execution Providers: https://onnxruntime.ai/docs/execution-providers
+
+        if cuda and providers is not None:
+            warnings.warn(
+                f"`cuda` and `providers` are mutually exclusive parameters, cuda: {cuda}, providers: {providers}",
+                category=UserWarning,
+                stacklevel=6,
+            )
+
+        if providers is not None:
+            onnx_providers = list(providers)
+        elif cuda:
+            if device_id is None:
+                onnx_providers = ["CUDAExecutionProvider"]
+            else:
+                onnx_providers = [("CUDAExecutionProvider", {"device_id": device_id})]
+        else:
+            onnx_providers = ["CPUExecutionProvider"]
+
+        available_providers = ort.get_available_providers()
+        requested_provider_names: list[str] = []
+        for provider in onnx_providers:
+            # check providers available
+            provider_name = provider if isinstance(provider, str) else provider[0]
+            requested_provider_names.append(provider_name)
+            if provider_name not in available_providers:
+                raise ValueError(
+                    f"Provider {provider_name} is not available. Available providers: {available_providers}"
+                )
+
+        so = ort.SessionOptions()
+        so.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL
+
+        if threads is not None:
+            so.intra_op_num_threads = threads
+            so.inter_op_num_threads = threads
+
+        self.model = ort.InferenceSession(
+            str(model_path), providers=onnx_providers, sess_options=so
+        )
+        if "CUDAExecutionProvider" in requested_provider_names:
+            assert self.model is not None
+            current_providers = self.model.get_providers()
+            if "CUDAExecutionProvider" not in current_providers:
+                warnings.warn(
+                    f"Attempt to set CUDAExecutionProvider failed. Current providers: {current_providers}."
+                    "If you are using CUDA 12.x, install onnxruntime-gpu via "
+                    "`pip install onnxruntime-gpu --extra-index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/simple/`",
+                    RuntimeWarning,
+                )
+
+    def load_onnx_model(self) -> None:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def onnx_embed(self, *args: Any, **kwargs: Any) -> OnnxOutputContext:
+        raise NotImplementedError("Subclasses must implement this method")
+
+
+class EmbeddingWorker(Worker, Generic[T]):
+    def init_embedding(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ) -> OnnxModel[T]:
+        raise NotImplementedError()
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ):
+        self.model = self.init_embedding(model_name, cache_dir, **kwargs)
+
+    @classmethod
+    def start(cls, model_name: str, cache_dir: str, **kwargs: Any) -> "EmbeddingWorker[T]":
+        return cls(model_name=model_name, cache_dir=cache_dir, **kwargs)
+
+    def process(self, items: Iterable[tuple[int, Any]]) -> Iterable[tuple[int, Any]]:
+        raise NotImplementedError("Subclasses must implement this method")
@@ -0,0 +1,83 @@
+import json
+from typing import Any
+from pathlib import Path
+
+from tokenizers import AddedToken, Tokenizer
+
+from fastembed.image.transform.operators import Compose
+
+
+def load_special_tokens(model_dir: Path) -> dict[str, Any]:
+    tokens_map_path = model_dir / "special_tokens_map.json"
+    if not tokens_map_path.exists():
+        raise ValueError(f"Could not find special_tokens_map.json in {model_dir}")
+
+    with open(str(tokens_map_path)) as tokens_map_file:
+        tokens_map = json.load(tokens_map_file)
+
+    return tokens_map
+
+
+def load_tokenizer(model_dir: Path) -> tuple[Tokenizer, dict[str, int]]:
+    config_path = model_dir / "config.json"
+    if not config_path.exists():
+        raise ValueError(f"Could not find config.json in {model_dir}")
+
+    tokenizer_path = model_dir / "tokenizer.json"
+    if not tokenizer_path.exists():
+        raise ValueError(f"Could not find tokenizer.json in {model_dir}")
+
+    tokenizer_config_path = model_dir / "tokenizer_config.json"
+    if not tokenizer_config_path.exists():
+        raise ValueError(f"Could not find tokenizer_config.json in {model_dir}")
+
+    with open(str(config_path)) as config_file:
+        config = json.load(config_file)
+
+    with open(str(tokenizer_config_path)) as tokenizer_config_file:
+        tokenizer_config = json.load(tokenizer_config_file)
+        assert (
+            "model_max_length" in tokenizer_config or "max_length" in tokenizer_config
+        ), "Models without model_max_length or max_length are not supported."
+        if "model_max_length" not in tokenizer_config:
+            max_context = tokenizer_config["max_length"]
+        elif "max_length" not in tokenizer_config:
+            max_context = tokenizer_config["model_max_length"]
+        else:
+            max_context = min(tokenizer_config["model_max_length"], tokenizer_config["max_length"])
+
+    tokens_map = load_special_tokens(model_dir)
+
+    tokenizer = Tokenizer.from_file(str(tokenizer_path))
+    tokenizer.enable_truncation(max_length=max_context)
+    tokenizer.enable_padding(
+        pad_id=config.get("pad_token_id", 0), pad_token=tokenizer_config["pad_token"]
+    )
+
+    for token in tokens_map.values():
+        if isinstance(token, str):
+            tokenizer.add_special_tokens([token])
+        elif isinstance(token, dict):
+            tokenizer.add_special_tokens([AddedToken(**token)])
+
+    special_token_to_id: dict[str, int] = {}
+
+    for token in tokens_map.values():
+        if isinstance(token, str):
+            special_token_to_id[token] = tokenizer.token_to_id(token)
+        elif isinstance(token, dict):
+            token_str = token.get("content", "")
+            special_token_to_id[token_str] = tokenizer.token_to_id(token_str)
+
+    return tokenizer, special_token_to_id
+
+
+def load_preprocessor(model_dir: Path) -> Compose:
+    preprocessor_config_path = model_dir / "preprocessor_config.json"
+    if not preprocessor_config_path.exists():
+        raise ValueError(f"Could not find preprocessor_config.json in {model_dir}")
+
+    with open(str(preprocessor_config_path)) as preprocessor_config_file:
+        preprocessor_config = json.load(preprocessor_config_file)
+        transforms = Compose.from_config(preprocessor_config)
+    return transforms
@@ -0,0 +1,24 @@
+from pathlib import Path
+import sys
+from PIL import Image
+from typing import Any, Union
+import numpy as np
+from numpy.typing import NDArray
+
+if sys.version_info >= (3, 10):
+    from typing import TypeAlias
+else:
+    from typing_extensions import TypeAlias
+
+
+PathInput: TypeAlias = Union[str, Path]
+ImageInput: TypeAlias = Union[PathInput, Image.Image]
+
+OnnxProvider: TypeAlias = Union[str, tuple[str, dict[Any, Any]]]
+NumpyArray = Union[
+    NDArray[np.float32],
+    NDArray[np.float16],
+    NDArray[np.int8],
+    NDArray[np.int64],
+    NDArray[np.int32],
+]
@@ -0,0 +1,69 @@
+import os
+import sys
+import re
+import tempfile
+import unicodedata
+from pathlib import Path
+from itertools import islice
+from typing import Iterable, Optional, TypeVar
+
+import numpy as np
+from numpy.typing import NDArray
+
+from fastembed.common.types import NumpyArray
+
+T = TypeVar("T")
+
+
+def normalize(input_array: NumpyArray, p: int = 2, dim: int = 1, eps: float = 1e-12) -> NumpyArray:
+    # Calculate the Lp norm along the specified dimension
+    norm = np.linalg.norm(input_array, ord=p, axis=dim, keepdims=True)
+    norm = np.maximum(norm, eps)  # Avoid division by zero
+    normalized_array = input_array / norm
+    return normalized_array
+
+
+def mean_pooling(input_array: NumpyArray, attention_mask: NDArray[np.int64]) -> NumpyArray:
+    input_mask_expanded = np.expand_dims(attention_mask, axis=-1).astype(np.int64)
+    input_mask_expanded = np.tile(input_mask_expanded, (1, 1, input_array.shape[-1]))
+    sum_embeddings = np.sum(input_array * input_mask_expanded, axis=1)
+    sum_mask = np.sum(input_mask_expanded, axis=1)
+    pooled_embeddings = sum_embeddings / np.maximum(sum_mask, 1e-9)
+    return pooled_embeddings
+
+
+def iter_batch(iterable: Iterable[T], size: int) -> Iterable[list[T]]:
+    """
+    >>> list(iter_batch([1,2,3,4,5], 3))
+    [[1, 2, 3], [4, 5]]
+    """
+    source_iter = iter(iterable)
+    while source_iter:
+        b = list(islice(source_iter, size))
+        if len(b) == 0:
+            break
+        yield b
+
+
+def define_cache_dir(cache_dir: Optional[str] = None) -> Path:
+    """
+    Define the cache directory for fastembed
+    """
+    if cache_dir is None:
+        default_cache_dir = os.path.join(tempfile.gettempdir(), "fastembed_cache")
+        cache_path = Path(os.getenv("FASTEMBED_CACHE_PATH", default_cache_dir))
+    else:
+        cache_path = Path(cache_dir)
+    cache_path.mkdir(parents=True, exist_ok=True)
+
+    return cache_path
+
+
+def get_all_punctuation() -> set[str]:
+    return set(
+        chr(i) for i in range(sys.maxunicode) if unicodedata.category(chr(i)).startswith("P")
+    )
+
+
+def remove_non_alphanumeric(text: str) -> str:
+    return re.sub(r"[^\w\s]", " ", text, flags=re.UNICODE)
@@ -1,309 +1,24 @@
-import os
-import shutil
-import tarfile
-from abc import ABC, abstractmethod
-from pathlib import Path
-from typing import Iterable, List
+from typing import Optional, Any

-import numpy as np
-import requests
-from optimum.onnxruntime import ORTModelForFeatureExtraction
-from tqdm import tqdm
-from transformers import AutoTokenizer
+from loguru import logger
+
+from fastembed import TextEmbedding
+
+logger.warning(
+    "DefaultEmbedding, FlagEmbedding, JinaEmbedding are deprecated."
+    "Use from fastembed import TextEmbedding instead."
+)
+
+DefaultEmbedding = TextEmbedding
+FlagEmbedding = TextEmbedding


-def normalize(input_array, p=2.0, dim=1, eps=1e-12):
-    # Calculate the Lp norm along the specified dimension
-    norm = np.linalg.norm(input_array, ord=p, axis=dim, keepdims=True)
-    norm = np.maximum(norm, eps) # Avoid division by zero
-    normalized_array = input_array / norm
-    return normalized_array
-
-class Embedding(ABC):
-    """
-    Abstract class for embeddings.
-
-    Args:
-        ABC ():
-
-    Raises:
-        NotImplementedError: Raised when you call an abstract method that has not been implemented.
-        PermissionError: _description_
-        ValueError: Several possible reasons: 1) targz_path does not exist or is not a file, 2) targz_path is not a .tar.gz file, 3) An error occurred while decompressing targz_path, 4) Could not find model_dir in cache_dir, 5) Could not find tokenizer.json in model_dir, 6) Could not find model.onnx in model_dir.
-        NotImplementedError: _description_
-
-    Returns:
-        _type_: _description_
-
-    Yields:
-        _type_: _description_
-    """
-
-    @abstractmethod
-    def embed(self, texts: List[str]) -> List[np.ndarray]:
-        raise NotImplementedError
-
-    @classmethod
-    def download_file_from_gcs(cls, url: str, output_path: str, show_progress: bool = True) -> str:
-        """
-        Downloads a file from Google Cloud Storage.
-
-        Args:
-            url (str): The URL to download the file from.
-            output_path (str): The path to save the downloaded file to.
-            show_progress (bool, optional): Whether to show a progress bar. Defaults to True.
-
-        Returns:
-            str: The path to the downloaded file.
-        """
-
-        if os.path.exists(output_path):
-            return output_path
-        response = requests.get(url, stream=True)
-
-        # Handle HTTP errors
-        if response.status_code == 403:
-            raise PermissionError(
-                "Authentication Error: You do not have permission to access this resource. Please check your credentials."
-            )
-
-        # Get the total size of the file
-        total_size_in_bytes = int(response.headers.get("content-length", 0))
-
-        # Warn if the total size is zero
-        if total_size_in_bytes == 0:
-            print(f"Warning: Content-length header is missing or zero in the response from {url}.")
-
-        # Initialize the progress bar
-        progress_bar = (
-            tqdm(total=total_size_in_bytes, unit="iB", unit_scale=True)
-            if total_size_in_bytes and show_progress
-            else None
-        )
-
-        # Attempt to download the file
-        try:
-            with open(output_path, "wb") as file:
-                for chunk in response.iter_content(chunk_size=1024):  # Adjust chunk size to your preference
-                    if chunk:  # Filter out keep-alive new chunks
-                        if progress_bar is not None:
-                            progress_bar.update(len(chunk))
-                        file.write(chunk)
-        except Exception as e:
-            print(f"An error occurred while trying to download the file: {str(e)}")
-            return
-        finally:
-            if progress_bar is not None:
-                progress_bar.close()
-        return output_path
-
-    @classmethod
-    def decompress_to_cache(cls, targz_path: str, cache_dir: str):
-        """
-        Decompresses a .tar.gz file to a cache directory.
-
-        Args:
-            targz_path (str): Path to the .tar.gz file.
-            cache_dir (str): Path to the cache directory.
-
-        Returns:
-            cache_dir (str): Path to the cache directory.
-        """
-        # Check if targz_path exists and is a file
-        if not os.path.isfile(targz_path):
-            raise ValueError(f"{targz_path} does not exist or is not a file.")
-
-        # Check if targz_path is a .tar.gz file
-        if not targz_path.endswith(".tar.gz"):
-            raise ValueError(f"{targz_path} is not a .tar.gz file.")
-
-        try:
-            # Open the tar.gz file
-            with tarfile.open(targz_path, "r:gz") as tar:
-                # Extract all files into the cache directory
-                tar.extractall(path=cache_dir)
-        except tarfile.TarError as e:
-            # If any error occurs while opening or extracting the tar.gz file,
-            # delete the cache directory (if it was created in this function)
-            # and raise the error again
-            if "tmp" in cache_dir:
-                shutil.rmtree(cache_dir)
-            raise ValueError(f"An error occurred while decompressing {targz_path}: {e}")
-
-        return cache_dir
-
-    def retrieve_model(self, model_name: str, cache_dir: str) -> Path:
-        """
-        Retrieves a model from Google Cloud Storage.
-
-        Args:
-            model_name (str): The name of the model to retrieve.
-            cache_dir (str): The path to the cache directory.
-
-        Raises:
-            ValueError: If the model_name is not in the format <org>/<model> e.g. BAAI/bge-base-en.
-
-        Returns:
-            Path: The path to the model directory.
-        """
-
-        assert "/" in model_name, "model_name must be in the format <org>/<model> e.g. BAAI/bge-base-en"
-
-        fast_model_name = f"fast-{model_name.split('/')[-1]}"
-
-        model_dir = Path(cache_dir) / fast_model_name
-        if model_dir.exists():
-            return model_dir
-
-        model_tar_gz = Path(cache_dir) / f"{fast_model_name}.tar.gz"
-        try:
-            self.download_file_from_gcs(
-                f"https://storage.googleapis.com/qdrant-fastembed/{fast_model_name}.tar.gz",
-            output_path=str(model_tar_gz),
-            )
-        except PermissionError:
-            simple_model_name = model_name.replace("/", "-")
-            print(f"Was not able to download {fast_model_name}.tar.gz, trying {simple_model_name}.tar.gz")
-            self.download_file_from_gcs(
-                f"https://storage.googleapis.com/qdrant-fastembed/{simple_model_name}.tar.gz",
-                output_path=str(model_tar_gz),
-            )
-
-        self.decompress_to_cache(targz_path=str(model_tar_gz), cache_dir=cache_dir)
-        assert model_dir.exists(), f"Could not find {model_dir} in {cache_dir}"
-
-        model_tar_gz.unlink()
-
-        return model_dir
-
-    def passage_embed(self, texts: List[str], batch_size: int = 256) -> Iterable[np.ndarray]:
-        """
-        Embeds a list of text passages into a list of embeddings.
-
-        Args:
-            texts (List[str]): The list of texts to embed.
-            batch_size (int, optional): The batch size. Defaults to 256.
-
-        Yields:
-            Iterable[np.ndarray]: The embeddings.
-        """
-
-        for i in range(0, len(texts), batch_size):
-            # Prepend "passage: " to each text
-            yield from self.embed([f"passage: {t}" for t in texts[i : i + batch_size]])
-
-    def query_embed(self, query: str) -> Iterable[np.ndarray]:
-        """
-        Embeds a query
-
-        Args:
-            query (str): The query to search for.
-
-        Returns:
-            Iterable[np.ndarray]: The embeddings.
-        """
-
-        # Prepend "query: " to the query
-        query = f"query: {query}"
-        # Embed the query
-        query_embedding = self.embed([query])
-        # Compute the cosine similarity between the query embedding and the document embeddings
-        return query_embedding
-
-
-class FlagEmbedding(Embedding):
-    """
-    Implementation of the Flag Embedding model.
-
-    Args:
-        Embedding (_type_): _description_
-    """
-
+class JinaEmbedding(TextEmbedding):
    def __init__(
        self,
-        model_name: str = "BAAI/bge-small-en",
-        max_length: int = 512,
-        cache_dir: str = None,
+        model_name: str = "jinaai/jina-embeddings-v2-base-en",
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        **kwargs: Any,
    ):
-        """
-        Args:
-            model_name (str): The name of the model to use.
-            max_length (int, optional): The maximum number of tokens. Defaults to 512. Unknown behavior for values > 512.
-
-        Raises:
-            ValueError: If the model_name is not in the format <org>/<model> e.g. BAAI/bge-base-en.
-        """
-        if cache_dir is None:
-            cache_dir = Path(".").resolve() / "local_cache"
-            cache_dir.mkdir(parents=True, exist_ok=True)
-        model_dir = self.retrieve_model(model_name, cache_dir)
-        if not (model_dir / "tokenizer.json").exists():
-            raise ValueError(f"Could not find tokenizer.json in {model_dir}")
-        if not (model_dir / "model.onnx").exists():
-            # Rename file model_optimized.onnx to model.onnx if it exists
-            if (model_dir / "model_optimized.onnx").exists():
-                (model_dir / "model_optimized.onnx").rename(model_dir / "model.onnx")
-            else:
-                raise ValueError(f"Could not find model.onnx in {model_dir}")
-
-        self.tokenizer = AutoTokenizer.from_pretrained(str(model_dir))
-        self.model = ORTModelForFeatureExtraction.from_pretrained(str(model_dir))
-
-    def onnx_embed(self, documents: List[str]) -> Iterable[np.ndarray]:
-        encoded_input = self.tokenizer(documents, padding=True, truncation=True, return_tensors='pt')
-        model_output = self.model(**encoded_input)
-        embeddings = model_output[0][:, 0]
-        return normalize(embeddings, p=2, dim=1)
-
-    def embed(self, documents: List[str], batch_size: int = 256) -> Iterable[np.ndarray]:
-        """
-        Encode a list of documents into list of embeddings.
-        We use mean pooling with attention so that the model can handle variable-length inputs.
-
-        Args:
-            documents: List of documents to embed
-            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
-
-        Returns:
-            List of embeddings, one per document
-        """
-        # TODO: Replace loop with parallelized batching
-        if len(documents) >= batch_size:
-            for i in range(0, len(documents), batch_size):
-                batch = documents[i : i + batch_size]
-                return self.onnx_embed(batch)
-        else:
-            return self.onnx_embed(documents)
-
-
-class DefaultEmbedding(FlagEmbedding):
-    """
-    Implementation of the default Flag Embedding model.
-
-    Args:
-        FlagEmbedding (_type_): _description_
-    """
-
-    def __init__(
-        self,
-        model_name: str = "BAAI/bge-small-en",
-        onnx_providers: List[str] = None,
-        max_length: int = 512,
-        cache_dir: str = None,
-    ):
-        # if onnx_providers is None:
-        #     onnx_providers = [ONNXProviders.CPU]
-        super().__init__(model_name, max_length=max_length, cache_dir=cache_dir)
-
-
-class OpenAIEmbedding(Embedding):
-    def __init__(self):
-        # Initialize your OpenAI model here
-        # self.model = ...
-        ...
-
-    def embed(self, texts):
-        # Use your OpenAI model to embed the texts
-        # return self.model.embed(texts)
-        raise NotImplementedError
-        raise NotImplementedError
+        super().__init__(model_name, cache_dir, threads, **kwargs)
@@ -0,0 +1,3 @@
+from fastembed.image.image_embedding import ImageEmbedding
+
+__all__ = ["ImageEmbedding"]
@@ -0,0 +1,101 @@
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+from dataclasses import asdict
+
+from fastembed.common.types import NumpyArray
+from fastembed.common import ImageInput, OnnxProvider
+from fastembed.image.image_embedding_base import ImageEmbeddingBase
+from fastembed.image.onnx_embedding import OnnxImageEmbedding
+from fastembed.common.model_description import DenseModelDescription
+
+
+class ImageEmbedding(ImageEmbeddingBase):
+    EMBEDDINGS_REGISTRY: list[Type[ImageEmbeddingBase]] = [OnnxImageEmbedding]
+
+    @classmethod
+    def list_supported_models(cls) -> list[dict[str, Any]]:
+        """
+        Lists the supported models.
+
+        Returns:
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
+
+            Example:
+                ```
+                [
+                    {
+                        "model": "Qdrant/clip-ViT-B-32-vision",
+                        "dim": 512,
+                        "description": "CLIP vision encoder based on ViT-B/32",
+                        "license": "mit",
+                        "size_in_GB": 0.33,
+                        "sources": {
+                            "hf": "Qdrant/clip-ViT-B-32-vision",
+                        },
+                        "model_file": "model.onnx",
+                    }
+                ]
+                ```
+        """
+        return [asdict(model) for model in cls._list_supported_models()]
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        result: list[DenseModelDescription] = []
+        for embedding in cls.EMBEDDINGS_REGISTRY:
+            result.extend(embedding._list_supported_models())
+        return result
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        **kwargs: Any,
+    ):
+        super().__init__(model_name, cache_dir, threads, **kwargs)
+        for EMBEDDING_MODEL_TYPE in self.EMBEDDINGS_REGISTRY:
+            supported_models = EMBEDDING_MODEL_TYPE._list_supported_models()
+            if any(model_name.lower() == model.model.lower() for model in supported_models):
+                self.model = EMBEDDING_MODEL_TYPE(
+                    model_name,
+                    cache_dir,
+                    threads=threads,
+                    providers=providers,
+                    cuda=cuda,
+                    device_ids=device_ids,
+                    lazy_load=lazy_load,
+                    **kwargs,
+                )
+                return
+
+        raise ValueError(
+            f"Model {model_name} is not supported in ImageEmbedding."
+            "Please check the supported models using `ImageEmbedding.list_supported_models()`"
+        )
+
+    def embed(
+        self,
+        images: Union[ImageInput, Iterable[ImageInput]],
+        batch_size: int = 16,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        """
+        Encode a list of images into list of embeddings.
+
+        Args:
+            images: Iterator of image paths or single image path to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per document
+        """
+        yield from self.model.embed(images, batch_size, parallel, **kwargs)
@@ -0,0 +1,44 @@
+from typing import Iterable, Optional, Any, Union
+
+from fastembed.common.model_description import DenseModelDescription
+from fastembed.common.types import NumpyArray
+from fastembed.common.model_management import ModelManagement
+from fastembed.common.types import ImageInput
+
+
+class ImageEmbeddingBase(ModelManagement[DenseModelDescription]):
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        **kwargs: Any,
+    ):
+        self.model_name = model_name
+        self.cache_dir = cache_dir
+        self.threads = threads
+        self._local_files_only = kwargs.pop("local_files_only", False)
+
+    def embed(
+        self,
+        images: Union[ImageInput, Iterable[ImageInput]],
+        batch_size: int = 16,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        """
+        Embeds a list of images into a list of embeddings.
+
+        Args:
+            images: The list of image paths to preprocess and embed.
+            batch_size: Batch size for encoding
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+            **kwargs: Additional keyword argument to pass to the embed method.
+
+        Yields:
+            Iterable[NdArray]: The embeddings.
+        """
+        raise NotImplementedError()
@@ -0,0 +1,207 @@
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+
+
+from fastembed.common.types import NumpyArray
+from fastembed.common import ImageInput, OnnxProvider
+from fastembed.common.onnx_model import OnnxOutputContext
+from fastembed.common.utils import define_cache_dir, normalize
+from fastembed.image.image_embedding_base import ImageEmbeddingBase
+from fastembed.image.onnx_image_model import ImageEmbeddingWorker, OnnxImageModel
+
+from fastembed.common.model_description import DenseModelDescription, ModelSource
+
+supported_onnx_models: list[DenseModelDescription] = [
+    DenseModelDescription(
+        model="Qdrant/clip-ViT-B-32-vision",
+        dim=512,
+        description="Image embeddings, Multimodal (text&image), 2021 year",
+        license="mit",
+        size_in_GB=0.34,
+        sources=ModelSource(hf="Qdrant/clip-ViT-B-32-vision"),
+        model_file="model.onnx",
+    ),
+    DenseModelDescription(
+        model="Qdrant/resnet50-onnx",
+        dim=2048,
+        description="Image embeddings, Unimodal (image), 2016 year",
+        license="apache-2.0",
+        size_in_GB=0.1,
+        sources=ModelSource(hf="Qdrant/resnet50-onnx"),
+        model_file="model.onnx",
+    ),
+    DenseModelDescription(
+        model="Qdrant/Unicom-ViT-B-16",
+        dim=768,
+        description="Image embeddings (more detailed than Unicom-ViT-B-32), Multimodal (text&image), 2023 year",
+        license="apache-2.0",
+        size_in_GB=0.82,
+        sources=ModelSource(hf="Qdrant/Unicom-ViT-B-16"),
+        model_file="model.onnx",
+    ),
+    DenseModelDescription(
+        model="Qdrant/Unicom-ViT-B-32",
+        dim=512,
+        description="Image embeddings, Multimodal (text&image), 2023 year",
+        license="apache-2.0",
+        size_in_GB=0.48,
+        sources=ModelSource(hf="Qdrant/Unicom-ViT-B-32"),
+        model_file="model.onnx",
+    ),
+    DenseModelDescription(
+        model="jinaai/jina-clip-v1",
+        dim=768,
+        description="Image embeddings, Multimodal (text&image), 2024 year",
+        license="apache-2.0",
+        size_in_GB=0.34,
+        sources=ModelSource(hf="jinaai/jina-clip-v1"),
+        model_file="onnx/vision_model.onnx",
+    ),
+]
+
+
+class OnnxImageEmbedding(ImageEmbeddingBase, OnnxImageModel[NumpyArray]):
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        device_id: Optional[int] = None,
+        specific_model_path: Optional[str] = None,
+        **kwargs: Any,
+    ):
+        """
+        Args:
+            model_name (str): The name of the model to use.
+            cache_dir (str, optional): The path to the cache directory.
+                                       Can be set using the `FASTEMBED_CACHE_PATH` env variable.
+                                       Defaults to `fastembed_cache` in the system's temp directory.
+            threads (int, optional): The number of threads single onnxruntime session can use. Defaults to None.
+            providers (Optional[Sequence[OnnxProvider]], optional): The list of onnxruntime providers to use.
+                Mutually exclusive with the `cuda` and `device_ids` arguments. Defaults to None.
+            cuda (bool, optional): Whether to use cuda for inference. Mutually exclusive with `providers`
+                Defaults to False.
+            device_ids (Optional[list[int]], optional): The list of device ids to use for data parallel processing in
+                workers. Should be used with `cuda=True`, mutually exclusive with `providers`. Defaults to None.
+            lazy_load (bool, optional): Whether to load the model during class initialization or on demand.
+                Should be set to True when using multiple-gpu and parallel encoding. Defaults to False.
+            device_id (Optional[int], optional): The device id to use for loading the model in the worker process.
+            specific_model_path (Optional[str], optional): The specific path to the onnx model dir if it should be imported from somewhere else
+
+        Raises:
+            ValueError: If the model_name is not in the format <org>/<model> e.g. BAAI/bge-base-en.
+        """
+
+        super().__init__(model_name, cache_dir, threads, **kwargs)
+        self.providers = providers
+        self.lazy_load = lazy_load
+
+        # List of device ids, that can be used for data parallel processing in workers
+        self.device_ids = device_ids
+        self.cuda = cuda
+
+        # This device_id will be used if we need to load model in current process
+        self.device_id: Optional[int] = None
+        if device_id is not None:
+            self.device_id = device_id
+        elif self.device_ids is not None:
+            self.device_id = self.device_ids[0]
+
+        self.model_description = self._get_model_description(model_name)
+        self.cache_dir = str(define_cache_dir(cache_dir))
+        self._model_dir = self.download_model(
+            self.model_description,
+            self.cache_dir,
+            local_files_only=self._local_files_only,
+            specific_model_path=specific_model_path,
+        )
+
+        if not self.lazy_load:
+            self.load_onnx_model()
+
+    def load_onnx_model(self) -> None:
+        """
+        Load the onnx model.
+        """
+        self._load_onnx_model(
+            model_dir=self._model_dir,
+            model_file=self.model_description.model_file,
+            threads=self.threads,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_id=self.device_id,
+        )
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        """
+        Lists the supported models.
+
+        Returns:
+            list[DenseModelDescription]: A list of DenseModelDescription objects containing the model information.
+        """
+        return supported_onnx_models
+
+    def embed(
+        self,
+        images: Union[ImageInput, Iterable[ImageInput]],
+        batch_size: int = 16,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        """
+        Encode a list of images into list of embeddings.
+        We use mean pooling with attention so that the model can handle variable-length inputs.
+
+        Args:
+            images: Iterator of image paths or single image path to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per document
+        """
+
+        yield from self._embed_images(
+            model_name=self.model_name,
+            cache_dir=str(self.cache_dir),
+            images=images,
+            batch_size=batch_size,
+            parallel=parallel,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_ids=self.device_ids,
+            **kwargs,
+        )
+
+    @classmethod
+    def _get_worker_class(cls) -> Type["ImageEmbeddingWorker[NumpyArray]"]:
+        return OnnxImageEmbeddingWorker
+
+    def _preprocess_onnx_input(
+        self, onnx_input: dict[str, NumpyArray], **kwargs: Any
+    ) -> dict[str, NumpyArray]:
+        """
+        Preprocess the onnx input.
+        """
+
+        return onnx_input
+
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[NumpyArray]:
+        return normalize(output.model_output)
+
+
+class OnnxImageEmbeddingWorker(ImageEmbeddingWorker[NumpyArray]):
+    def init_embedding(self, model_name: str, cache_dir: str, **kwargs: Any) -> OnnxImageEmbedding:
+        return OnnxImageEmbedding(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )
@@ -0,0 +1,135 @@
+import contextlib
+import os
+from multiprocessing import get_all_start_methods
+from pathlib import Path
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+
+import numpy as np
+from PIL import Image
+
+from fastembed.image.transform.operators import Compose
+from fastembed.common.types import NumpyArray
+from fastembed.common import ImageInput, OnnxProvider
+from fastembed.common.onnx_model import EmbeddingWorker, OnnxModel, OnnxOutputContext, T
+from fastembed.common.preprocessor_utils import load_preprocessor
+from fastembed.common.utils import iter_batch
+from fastembed.parallel_processor import ParallelWorkerPool
+
+# Holds type of the embedding result
+
+
+class OnnxImageModel(OnnxModel[T]):
+    @classmethod
+    def _get_worker_class(cls) -> Type["ImageEmbeddingWorker[T]"]:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[T]:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def __init__(self) -> None:
+        super().__init__()
+        self.processor: Optional[Compose] = None
+
+    def _preprocess_onnx_input(
+        self, onnx_input: dict[str, NumpyArray], **kwargs: Any
+    ) -> dict[str, NumpyArray]:
+        """
+        Preprocess the onnx input.
+        """
+        return onnx_input
+
+    def _load_onnx_model(
+        self,
+        model_dir: Path,
+        model_file: str,
+        threads: Optional[int],
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_id: Optional[int] = None,
+    ) -> None:
+        super()._load_onnx_model(
+            model_dir=model_dir,
+            model_file=model_file,
+            threads=threads,
+            providers=providers,
+            cuda=cuda,
+            device_id=device_id,
+        )
+        self.processor = load_preprocessor(model_dir=model_dir)
+
+    def load_onnx_model(self) -> None:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def _build_onnx_input(self, encoded: NumpyArray) -> dict[str, NumpyArray]:
+        input_name = self.model.get_inputs()[0].name  # type: ignore[union-attr]
+        return {input_name: encoded}
+
+    def onnx_embed(self, images: list[ImageInput], **kwargs: Any) -> OnnxOutputContext:
+        with contextlib.ExitStack():
+            image_files = [
+                Image.open(image) if not isinstance(image, Image.Image) else image
+                for image in images
+            ]
+            assert self.processor is not None, "Processor is not initialized"
+            encoded = np.array(self.processor(image_files))
+        onnx_input = self._build_onnx_input(encoded)
+        onnx_input = self._preprocess_onnx_input(onnx_input)
+        model_output = self.model.run(None, onnx_input)  # type: ignore[union-attr]
+        embeddings = model_output[0].reshape(len(images), -1)
+        return OnnxOutputContext(model_output=embeddings)
+
+    def _embed_images(
+        self,
+        model_name: str,
+        cache_dir: str,
+        images: Union[ImageInput, Iterable[ImageInput]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        **kwargs: Any,
+    ) -> Iterable[T]:
+        is_small = False
+
+        if isinstance(images, (str, Path, Image.Image)):
+            images = [images]
+            is_small = True
+
+        if isinstance(images, list) and len(images) < batch_size:
+            is_small = True
+
+        if parallel is None or is_small:
+            if not hasattr(self, "model") or self.model is None:
+                self.load_onnx_model()
+
+            for batch in iter_batch(images, batch_size):
+                yield from self._post_process_onnx_output(self.onnx_embed(batch))
+        else:
+            if parallel == 0:
+                parallel = os.cpu_count()
+
+            start_method = "forkserver" if "forkserver" in get_all_start_methods() else "spawn"
+            params = {
+                "model_name": model_name,
+                "cache_dir": cache_dir,
+                "providers": providers,
+                **kwargs,
+            }
+
+            pool = ParallelWorkerPool(
+                num_workers=parallel or 1,
+                worker=self._get_worker_class(),
+                cuda=cuda,
+                device_ids=device_ids,
+                start_method=start_method,
+            )
+            for batch in pool.ordered_map(iter_batch(images, batch_size), **params):
+                yield from self._post_process_onnx_output(batch)  # type: ignore
+
+
+class ImageEmbeddingWorker(EmbeddingWorker[T]):
+    def process(self, items: Iterable[tuple[int, Any]]) -> Iterable[tuple[int, Any]]:
+        for idx, batch in items:
+            embeddings = self.model.onnx_embed(batch)
+            yield idx, embeddings
@@ -0,0 +1,149 @@
+from typing import Union
+
+import numpy as np
+from PIL import Image
+
+from fastembed.common.types import NumpyArray
+
+
+def convert_to_rgb(image: Image.Image) -> Image.Image:
+    if image.mode == "RGB":
+        return image
+
+    image = image.convert("RGB")
+    return image
+
+
+def center_crop(
+    image: Union[Image.Image, NumpyArray],
+    size: tuple[int, int],
+) -> NumpyArray:
+    if isinstance(image, np.ndarray):
+        _, orig_height, orig_width = image.shape
+    else:
+        orig_height, orig_width = image.height, image.width
+        # (H, W, C) -> (C, H, W)
+        image = np.array(image).transpose((2, 0, 1))
+
+    crop_height, crop_width = size
+
+    # left upper corner (0, 0)
+    top = (orig_height - crop_height) // 2
+    bottom = top + crop_height
+    left = (orig_width - crop_width) // 2
+    right = left + crop_width
+
+    # Check if cropped area is within image boundaries
+    if top >= 0 and bottom <= orig_height and left >= 0 and right <= orig_width:
+        image = image[..., top:bottom, left:right]
+        return image
+
+    # Padding with zeros
+    new_height = max(crop_height, orig_height)
+    new_width = max(crop_width, orig_width)
+    new_shape = image.shape[:-2] + (new_height, new_width)
+    new_image = np.zeros_like(image, shape=new_shape, dtype=np.float32)
+
+    top_pad = (new_height - orig_height) // 2
+    bottom_pad = top_pad + orig_height
+    left_pad = (new_width - orig_width) // 2
+    right_pad = left_pad + orig_width
+    new_image[..., top_pad:bottom_pad, left_pad:right_pad] = image
+
+    top += top_pad
+    bottom += top_pad
+    left += left_pad
+    right += left_pad
+
+    new_image = new_image[
+        ..., max(0, top) : min(new_height, bottom), max(0, left) : min(new_width, right)
+    ]
+
+    return new_image
+
+
+def normalize(
+    image: NumpyArray,
+    mean: Union[float, list[float]],
+    std: Union[float, list[float]],
+) -> NumpyArray:
+    num_channels = image.shape[1] if len(image.shape) == 4 else image.shape[0]
+
+    if not np.issubdtype(image.dtype, np.floating):
+        image = image.astype(np.float32)
+
+    mean = mean if isinstance(mean, list) else [mean] * num_channels
+
+    if len(mean) != num_channels:
+        raise ValueError(
+            f"mean must have the same number of channels as the image, image has {num_channels} channels, got "
+            f"{len(mean)}"
+        )
+
+    mean_arr = np.array(mean, dtype=np.float32)
+
+    std = std if isinstance(std, list) else [std] * num_channels
+    if len(std) != num_channels:
+        raise ValueError(
+            f"std must have the same number of channels as the image, image has {num_channels} channels, got {len(std)}"
+        )
+
+    std_arr = np.array(std, dtype=np.float32)
+
+    image = ((image.T - mean_arr) / std_arr).T
+    return image
+
+
+def resize(
+    image: Image.Image,
+    size: Union[int, tuple[int, int]],
+    resample: Union[int, Image.Resampling] = Image.Resampling.BILINEAR,
+) -> Image.Image:
+    if isinstance(size, tuple):
+        return image.resize(size, resample)
+
+    height, width = image.height, image.width
+    short, long = (width, height) if width <= height else (height, width)
+
+    new_short, new_long = size, int(size * long / short)
+    if width <= height:
+        new_size = (new_short, new_long)
+    else:
+        new_size = (new_long, new_short)
+    return image.resize(new_size, resample)
+
+
+def rescale(image: NumpyArray, scale: float, dtype: type = np.float32) -> NumpyArray:
+    return (image * scale).astype(dtype)
+
+
+def pil2ndarray(image: Union[Image.Image, NumpyArray]) -> NumpyArray:
+    if isinstance(image, Image.Image):
+        return np.asarray(image).transpose((2, 0, 1))
+    return image
+
+
+def pad2square(
+    image: Image.Image,
+    size: int,
+    fill_color: Union[str, int, tuple[int, ...]] = 0,
+) -> Image.Image:
+    height, width = image.height, image.width
+
+    left, right = 0, width
+    top, bottom = 0, height
+
+    crop_required = False
+    if width > size:
+        left = (width - size) // 2
+        right = left + size
+        crop_required = True
+
+    if height > size:
+        top = (height - size) // 2
+        bottom = top + size
+        crop_required = True
+
+    new_image = Image.new(mode="RGB", size=(size, size), color=fill_color)
+    new_image.paste(image.crop((left, top, right, bottom)) if crop_required else image)
+    return new_image
@@ -0,0 +1,269 @@
+from typing import Any, Union, Optional
+
+from PIL import Image
+
+from fastembed.common.types import NumpyArray
+from fastembed.image.transform.functional import (
+    center_crop,
+    convert_to_rgb,
+    normalize,
+    pil2ndarray,
+    rescale,
+    resize,
+    pad2square,
+)
+
+
+class Transform:
+    def __call__(self, images: list[Any]) -> Union[list[Image.Image], list[NumpyArray]]:
+        raise NotImplementedError("Subclasses must implement this method")
+
+
+class ConvertToRGB(Transform):
+    def __call__(self, images: list[Image.Image]) -> list[Image.Image]:
+        return [convert_to_rgb(image=image) for image in images]
+
+
+class CenterCrop(Transform):
+    def __init__(self, size: tuple[int, int]):
+        self.size = size
+
+    def __call__(self, images: list[Image.Image]) -> list[NumpyArray]:
+        return [center_crop(image=image, size=self.size) for image in images]
+
+
+class Normalize(Transform):
+    def __init__(self, mean: Union[float, list[float]], std: Union[float, list[float]]):
+        self.mean = mean
+        self.std = std
+
+    def __call__(self, images: list[NumpyArray]) -> list[NumpyArray]:
+        return [normalize(image, mean=self.mean, std=self.std) for image in images]
+
+
+class Resize(Transform):
+    def __init__(
+        self,
+        size: Union[int, tuple[int, int]],
+        resample: Image.Resampling = Image.Resampling.BICUBIC,
+    ):
+        self.size = size
+        self.resample = resample
+
+    def __call__(self, images: list[Image.Image]) -> list[Image.Image]:
+        return [resize(image, size=self.size, resample=self.resample) for image in images]
+
+
+class Rescale(Transform):
+    def __init__(self, scale: float = 1 / 255):
+        self.scale = scale
+
+    def __call__(self, images: list[NumpyArray]) -> list[NumpyArray]:
+        return [rescale(image, scale=self.scale) for image in images]
+
+
+class PILtoNDarray(Transform):
+    def __call__(self, images: list[Union[Image.Image, NumpyArray]]) -> list[NumpyArray]:
+        return [pil2ndarray(image) for image in images]
+
+
+class PadtoSquare(Transform):
+    def __init__(
+        self,
+        size: int,
+        fill_color: Union[str, int, tuple[int, ...]],
+    ):
+        self.size = size
+        self.fill_color = fill_color
+
+    def __call__(self, images: list[Image.Image]) -> list[Image.Image]:
+        return [
+            pad2square(image=image, size=self.size, fill_color=self.fill_color) for image in images
+        ]
+
+
+class Compose:
+    def __init__(self, transforms: list[Transform]):
+        self.transforms = transforms
+
+    def __call__(
+        self, images: Union[list[Image.Image], list[NumpyArray]]
+    ) -> Union[list[NumpyArray], list[Image.Image]]:
+        for transform in self.transforms:
+            images = transform(images)
+        return images
+
+    @classmethod
+    def from_config(cls, config: dict[str, Any]) -> "Compose":
+        """Creates processor from a config dict.
+        Args:
+            config (dict[str, Any]): Configuration dictionary.
+
+                Valid keys:
+                    - do_resize
+                    - resize_mode
+                    - size
+                    - fill_color
+                    - do_center_crop
+                    - crop_size
+                    - do_rescale
+                    - rescale_factor
+                    - do_normalize
+                    - image_mean
+                    - mean
+                    - image_std
+                    - std
+                    - resample
+                    - interpolation
+                Valid size keys (nested):
+                    - {"height", "width"}
+                    - {"shortest_edge"}
+
+        Returns:
+            Compose: Image processor.
+        """
+        transforms: list[Transform] = []
+        cls._get_convert_to_rgb(transforms, config)
+        cls._get_resize(transforms, config)
+        cls._get_pad2square(transforms, config)
+        cls._get_center_crop(transforms, config)
+        cls._get_pil2ndarray(transforms, config)
+        cls._get_rescale(transforms, config)
+        cls._get_normalize(transforms, config)
+        return cls(transforms=transforms)
+
+    @staticmethod
+    def _get_convert_to_rgb(transforms: list[Transform], config: dict[str, Any]) -> None:
+        transforms.append(ConvertToRGB())
+
+    @classmethod
+    def _get_resize(cls, transforms: list[Transform], config: dict[str, Any]) -> None:
+        mode = config.get("image_processor_type", "CLIPImageProcessor")
+        if mode in ("CLIPImageProcessor", "SiglipImageProcessor"):
+            if config.get("do_resize", False):
+                size = config["size"]
+                if "shortest_edge" in size:
+                    size = size["shortest_edge"]
+                elif "height" in size and "width" in size:
+                    size = (size["height"], size["width"])
+                else:
+                    raise ValueError(
+                        "Size must contain either 'shortest_edge' or 'height' and 'width'."
+                    )
+                transforms.append(
+                    Resize(
+                        size=size,
+                        resample=config.get("resample", Image.Resampling.BICUBIC),
+                    )
+                )
+        elif mode == "ConvNextFeatureExtractor":
+            if "size" in config and "shortest_edge" not in config["size"]:
+                raise ValueError(
+                    f"Size dictionary must contain 'shortest_edge' key. Got {config['size'].keys()}"
+                )
+            shortest_edge = config["size"]["shortest_edge"]
+            crop_pct = config.get("crop_pct", 0.875)
+            if shortest_edge < 384:
+                # maintain same ratio, resizing shortest edge to shortest_edge/crop_pct
+                resize_shortest_edge = int(shortest_edge / crop_pct)
+                transforms.append(
+                    Resize(
+                        size=resize_shortest_edge,
+                        resample=config.get("resample", Image.Resampling.BICUBIC),
+                    )
+                )
+                transforms.append(CenterCrop(size=(shortest_edge, shortest_edge)))
+            else:
+                transforms.append(
+                    Resize(
+                        size=(shortest_edge, shortest_edge),
+                        resample=config.get("resample", Image.Resampling.BICUBIC),
+                    )
+                )
+        elif mode == "JinaCLIPImageProcessor":
+            interpolation = config.get("interpolation")
+            if isinstance(interpolation, str):
+                resample = cls._interpolation_resolver(interpolation)
+            else:
+                resample = interpolation or Image.Resampling.BICUBIC
+
+            if "size" in config:
+                resize_mode = config.get("resize_mode", "shortest")
+                if resize_mode == "shortest":
+                    transforms.append(
+                        Resize(
+                            size=config["size"],
+                            resample=resample,
+                        )
+                    )
+        else:
+            raise ValueError(f"Preprocessor {mode} is not supported")
+
+    @staticmethod
+    def _get_center_crop(transforms: list[Transform], config: dict[str, Any]) -> None:
+        mode = config.get("image_processor_type", "CLIPImageProcessor")
+        if mode in ("CLIPImageProcessor", "SiglipImageProcessor"):
+            if config.get("do_center_crop", False):
+                crop_size_raw = config["crop_size"]
+                crop_size: tuple[int, int]
+                if isinstance(crop_size_raw, int):
+                    crop_size = (crop_size_raw, crop_size_raw)
+                elif isinstance(crop_size_raw, dict):
+                    crop_size = (crop_size_raw["height"], crop_size_raw["width"])
+                else:
+                    raise ValueError(f"Invalid crop size: {crop_size_raw}")
+                transforms.append(CenterCrop(size=crop_size))
+        elif mode == "ConvNextFeatureExtractor":
+            pass
+        elif mode == "JinaCLIPImageProcessor":
+            pass
+        else:
+            raise ValueError(f"Preprocessor {mode} is not supported")
+
+    @staticmethod
+    def _get_pil2ndarray(transforms: list[Transform], config: dict[str, Any]) -> None:
+        transforms.append(PILtoNDarray())
+
+    @staticmethod
+    def _get_rescale(transforms: list[Transform], config: dict[str, Any]) -> None:
+        if config.get("do_rescale", True):
+            rescale_factor = config.get("rescale_factor", 1 / 255)
+            transforms.append(Rescale(scale=rescale_factor))
+
+    @staticmethod
+    def _get_normalize(transforms: list[Transform], config: dict[str, Any]) -> None:
+        if config.get("do_normalize", False):
+            transforms.append(Normalize(mean=config["image_mean"], std=config["image_std"]))
+        elif "mean" in config and "std" in config:
+            transforms.append(Normalize(mean=config["mean"], std=config["std"]))
+
+    @staticmethod
+    def _get_pad2square(transforms: list[Transform], config: dict[str, Any]) -> None:
+        mode = config.get("image_processor_type", "CLIPImageProcessor")
+        if mode == "CLIPImageProcessor":
+            pass
+        elif mode == "ConvNextFeatureExtractor":
+            pass
+        elif mode == "JinaCLIPImageProcessor":
+            transforms.append(
+                PadtoSquare(
+                    size=config["size"],
+                    fill_color=config.get("fill_color", 0),
+                )
+            )
+
+    @staticmethod
+    def _interpolation_resolver(resample: Optional[str] = None) -> Image.Resampling:
+        interpolation_map = {
+            "nearest": Image.Resampling.NEAREST,
+            "lanczos": Image.Resampling.LANCZOS,
+            "bilinear": Image.Resampling.BILINEAR,
+            "bicubic": Image.Resampling.BICUBIC,
+            "box": Image.Resampling.BOX,
+            "hamming": Image.Resampling.HAMMING,
+        }
+
+        if resample and (method := interpolation_map.get(resample.lower())):
+            return method
+
+        raise ValueError(f"Unknown interpolation method: {resample}")
@@ -0,0 +1,5 @@
+from fastembed.late_interaction.late_interaction_text_embedding import (
+    LateInteractionTextEmbedding,
+)
+
+__all__ = ["LateInteractionTextEmbedding"]
@@ -0,0 +1,263 @@
+import string
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+
+import numpy as np
+from tokenizers import Encoding
+
+from fastembed.common.types import NumpyArray
+from fastembed.common import OnnxProvider
+from fastembed.common.onnx_model import OnnxOutputContext
+from fastembed.common.utils import define_cache_dir
+from fastembed.late_interaction.late_interaction_embedding_base import (
+    LateInteractionTextEmbeddingBase,
+)
+from fastembed.text.onnx_text_model import OnnxTextModel, TextEmbeddingWorker
+from fastembed.common.model_description import DenseModelDescription, ModelSource
+
+supported_colbert_models: list[DenseModelDescription] = [
+    DenseModelDescription(
+        model="colbert-ir/colbertv2.0",
+        dim=128,
+        description="Late interaction model",
+        license="mit",
+        size_in_GB=0.44,
+        sources=ModelSource(hf="colbert-ir/colbertv2.0"),
+        model_file="model.onnx",
+    ),
+    DenseModelDescription(
+        model="answerdotai/answerai-colbert-small-v1",
+        dim=96,
+        description="Text embeddings, Unimodal (text), Multilingual (~100 languages), 512 input tokens truncation, 2024 year",
+        license="apache-2.0",
+        size_in_GB=0.13,
+        sources=ModelSource(hf="answerdotai/answerai-colbert-small-v1"),
+        model_file="vespa_colbert.onnx",
+    ),
+]
+
+
+class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[NumpyArray]):
+    QUERY_MARKER_TOKEN_ID = 1
+    DOCUMENT_MARKER_TOKEN_ID = 2
+    MIN_QUERY_LENGTH = 31  # it's 32, we add one additional special token in the beginning
+    MASK_TOKEN = "[MASK]"
+
+    def _post_process_onnx_output(
+        self, output: OnnxOutputContext, is_doc: bool = True
+    ) -> Iterable[NumpyArray]:
+        if not is_doc:
+            return output.model_output
+
+        if output.input_ids is None or output.attention_mask is None:
+            raise ValueError(
+                "input_ids and attention_mask must be provided for document post-processing"
+            )
+
+        for i, token_sequence in enumerate(output.input_ids):
+            for j, token_id in enumerate(token_sequence):  # type: ignore
+                if token_id in self.skip_list or token_id == self.pad_token_id:
+                    output.attention_mask[i, j] = 0
+
+        output.model_output *= np.expand_dims(output.attention_mask, 2)
+        norm = np.linalg.norm(output.model_output, ord=2, axis=2, keepdims=True)
+        norm_clamped = np.maximum(norm, 1e-12)
+        output.model_output /= norm_clamped
+        return output.model_output
+
+    def _preprocess_onnx_input(
+        self, onnx_input: dict[str, NumpyArray], is_doc: bool = True, **kwargs: Any
+    ) -> dict[str, NumpyArray]:
+        marker_token = self.DOCUMENT_MARKER_TOKEN_ID if is_doc else self.QUERY_MARKER_TOKEN_ID
+        onnx_input["input_ids"] = np.insert(
+            onnx_input["input_ids"].astype(np.int64), 1, marker_token, axis=1
+        )
+        onnx_input["attention_mask"] = np.insert(
+            onnx_input["attention_mask"].astype(np.int64), 1, 1, axis=1
+        )
+        return onnx_input
+
+    def tokenize(self, documents: list[str], is_doc: bool = True, **kwargs: Any) -> list[Encoding]:
+        return (
+            self._tokenize_documents(documents=documents)
+            if is_doc
+            else self._tokenize_query(query=next(iter(documents)))
+        )
+
+    def _tokenize_query(self, query: str) -> list[Encoding]:
+        assert self.tokenizer is not None
+        encoded = self.tokenizer.encode_batch([query])
+        # colbert authors recommend to pad queries with [MASK] tokens for query augmentation to improve performance
+        if len(encoded[0].ids) < self.MIN_QUERY_LENGTH:
+            prev_padding = None
+            if self.tokenizer.padding:
+                prev_padding = self.tokenizer.padding
+            self.tokenizer.enable_padding(
+                pad_token=self.MASK_TOKEN,
+                pad_id=self.mask_token_id,
+                length=self.MIN_QUERY_LENGTH,
+            )
+            encoded = self.tokenizer.encode_batch([query])
+            if prev_padding is None:
+                self.tokenizer.no_padding()
+            else:
+                self.tokenizer.enable_padding(**prev_padding)
+        return encoded
+
+    def _tokenize_documents(self, documents: list[str]) -> list[Encoding]:
+        encoded = self.tokenizer.encode_batch(documents)  # type: ignore[union-attr]
+        return encoded
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        """Lists the supported models.
+
+        Returns:
+            list[DenseModelDescription]: A list of DenseModelDescription objects containing the model information.
+        """
+        return supported_colbert_models
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        device_id: Optional[int] = None,
+        specific_model_path: Optional[str] = None,
+        **kwargs: Any,
+    ):
+        """
+        Args:
+            model_name (str): The name of the model to use.
+            cache_dir (str, optional): The path to the cache directory.
+                                       Can be set using the `FASTEMBED_CACHE_PATH` env variable.
+                                       Defaults to `fastembed_cache` in the system's temp directory.
+            threads (int, optional): The number of threads single onnxruntime session can use. Defaults to None.
+            providers (Optional[Sequence[OnnxProvider]], optional): The list of onnxruntime providers to use.
+                Mutually exclusive with the `cuda` and `device_ids` arguments. Defaults to None.
+            cuda (bool, optional): Whether to use cuda for inference. Mutually exclusive with `providers`
+                Defaults to False.
+            device_ids (Optional[list[int]], optional): The list of device ids to use for data parallel processing in
+                workers. Should be used with `cuda=True`, mutually exclusive with `providers`. Defaults to None.
+            lazy_load (bool, optional): Whether to load the model during class initialization or on demand.
+                Should be set to True when using multiple-gpu and parallel encoding. Defaults to False.
+            device_id (Optional[int], optional): The device id to use for loading the model in the worker process.
+            specific_model_path (Optional[str], optional): The specific path to the onnx model dir if it should be imported from somewhere else
+
+        Raises:
+            ValueError: If the model_name is not in the format <org>/<model> e.g. BAAI/bge-base-en.
+        """
+
+        super().__init__(model_name, cache_dir, threads, **kwargs)
+        self.providers = providers
+        self.lazy_load = lazy_load
+
+        # List of device ids, that can be used for data parallel processing in workers
+        self.device_ids = device_ids
+        self.cuda = cuda
+
+        # This device_id will be used if we need to load model in current process
+        self.device_id: Optional[int] = None
+        if device_id is not None:
+            self.device_id = device_id
+        elif self.device_ids is not None:
+            self.device_id = self.device_ids[0]
+
+        self.model_description = self._get_model_description(model_name)
+        self.cache_dir = str(define_cache_dir(cache_dir))
+
+        self._model_dir = self.download_model(
+            self.model_description,
+            self.cache_dir,
+            local_files_only=self._local_files_only,
+            specific_model_path=specific_model_path,
+        )
+        self.mask_token_id: Optional[int] = None
+        self.pad_token_id: Optional[int] = None
+        self.skip_list: set[int] = set()
+
+        if not self.lazy_load:
+            self.load_onnx_model()
+
+    def load_onnx_model(self) -> None:
+        self._load_onnx_model(
+            model_dir=self._model_dir,
+            model_file=self.model_description.model_file,
+            threads=self.threads,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_id=self.device_id,
+        )
+        assert self.tokenizer is not None
+        self.mask_token_id = self.special_token_to_id[self.MASK_TOKEN]
+        self.pad_token_id = self.tokenizer.padding["pad_id"]
+        self.skip_list = {
+            self.tokenizer.encode(symbol, add_special_tokens=False).ids[0]
+            for symbol in string.punctuation
+        }
+        current_max_length = self.tokenizer.truncation["max_length"]
+        # ensure not to overflow after adding document-marker
+        self.tokenizer.enable_truncation(max_length=current_max_length - 1)
+
+    def embed(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        """
+        Encode a list of documents into list of embeddings.
+        We use mean pooling with attention so that the model can handle variable-length inputs.
+
+        Args:
+            documents: Iterator of documents or single document to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per document
+        """
+        yield from self._embed_documents(
+            model_name=self.model_name,
+            cache_dir=str(self.cache_dir),
+            documents=documents,
+            batch_size=batch_size,
+            parallel=parallel,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_ids=self.device_ids,
+            **kwargs,
+        )
+
+    def query_embed(self, query: Union[str, Iterable[str]], **kwargs: Any) -> Iterable[NumpyArray]:
+        if isinstance(query, str):
+            query = [query]
+
+        if not hasattr(self, "model") or self.model is None:
+            self.load_onnx_model()
+
+        for text in query:
+            yield from self._post_process_onnx_output(
+                self.onnx_embed([text], is_doc=False), is_doc=False
+            )
+
+    @classmethod
+    def _get_worker_class(cls) -> Type[TextEmbeddingWorker[NumpyArray]]:
+        return ColbertEmbeddingWorker
+
+
+class ColbertEmbeddingWorker(TextEmbeddingWorker[NumpyArray]):
+    def init_embedding(self, model_name: str, cache_dir: str, **kwargs: Any) -> Colbert:
+        return Colbert(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )
@@ -0,0 +1,58 @@
+from typing import Any, Type
+
+from fastembed.common.types import NumpyArray
+from fastembed.late_interaction.colbert import Colbert, ColbertEmbeddingWorker
+from fastembed.common.model_description import DenseModelDescription, ModelSource
+
+supported_jina_colbert_models: list[DenseModelDescription] = [
+    DenseModelDescription(
+        model="jinaai/jina-colbert-v2",
+        dim=128,
+        description="New model that expands capabilities of colbert-v1 with multilingual and context length of 8192, 2024 year",
+        license="cc-by-nc-4.0",
+        size_in_GB=2.24,
+        sources=ModelSource(hf="jinaai/jina-colbert-v2"),
+        model_file="onnx/model.onnx",
+        additional_files=["onnx/model.onnx_data"],
+    )
+]
+
+
+class JinaColbert(Colbert):
+    QUERY_MARKER_TOKEN_ID = 250002
+    DOCUMENT_MARKER_TOKEN_ID = 250003
+    MIN_QUERY_LENGTH = 31  # it's 32, we add one additional special token in the beginning
+    MASK_TOKEN = "<mask>"
+
+    @classmethod
+    def _get_worker_class(cls) -> Type[ColbertEmbeddingWorker]:
+        return JinaColbertEmbeddingWorker
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        """Lists the supported models.
+
+        Returns:
+            list[DenseModelDescription]: A list of DenseModelDescription objects containing the model information.
+        """
+        return supported_jina_colbert_models
+
+    def _preprocess_onnx_input(
+        self, onnx_input: dict[str, NumpyArray], is_doc: bool = True, **kwargs: Any
+    ) -> dict[str, NumpyArray]:
+        onnx_input = super()._preprocess_onnx_input(onnx_input, is_doc)
+
+        # the attention mask for jina-colbert-v2 is always 1 in queries
+        if not is_doc:
+            onnx_input["attention_mask"][:] = 1
+        return onnx_input
+
+
+class JinaColbertEmbeddingWorker(ColbertEmbeddingWorker):
+    def init_embedding(self, model_name: str, cache_dir: str, **kwargs: Any) -> JinaColbert:
+        return JinaColbert(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )
@@ -0,0 +1,60 @@
+from typing import Iterable, Optional, Union, Any
+
+from fastembed.common.model_description import DenseModelDescription
+from fastembed.common.types import NumpyArray
+from fastembed.common.model_management import ModelManagement
+
+
+class LateInteractionTextEmbeddingBase(ModelManagement[DenseModelDescription]):
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        **kwargs: Any,
+    ):
+        self.model_name = model_name
+        self.cache_dir = cache_dir
+        self.threads = threads
+        self._local_files_only = kwargs.pop("local_files_only", False)
+
+    def embed(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        raise NotImplementedError()
+
+    def passage_embed(self, texts: Iterable[str], **kwargs: Any) -> Iterable[NumpyArray]:
+        """
+        Embeds a list of text passages into a list of embeddings.
+
+        Args:
+            texts (Iterable[str]): The list of texts to embed.
+            **kwargs: Additional keyword argument to pass to the embed method.
+
+        Yields:
+            Iterable[NdArray]: The embeddings.
+        """
+
+        # This is model-specific, so that different models can have specialized implementations
+        yield from self.embed(texts, **kwargs)
+
+    def query_embed(self, query: Union[str, Iterable[str]], **kwargs: Any) -> Iterable[NumpyArray]:
+        """
+        Embeds queries
+
+        Args:
+            query (Union[str, Iterable[str]]): The query to embed, or an iterable e.g. list of queries.
+
+        Returns:
+            Iterable[NdArray]: The embeddings.
+        """
+
+        # This is model-specific, so that different models can have specialized implementations
+        if isinstance(query, str):
+            yield from self.embed([query], **kwargs)
+        else:
+            yield from self.embed(query, **kwargs)
@@ -0,0 +1,119 @@
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+from dataclasses import asdict
+
+from fastembed.common.model_description import DenseModelDescription
+from fastembed.common.types import NumpyArray
+from fastembed.common import OnnxProvider
+from fastembed.late_interaction.colbert import Colbert
+from fastembed.late_interaction.jina_colbert import JinaColbert
+from fastembed.late_interaction.late_interaction_embedding_base import (
+    LateInteractionTextEmbeddingBase,
+)
+
+
+class LateInteractionTextEmbedding(LateInteractionTextEmbeddingBase):
+    EMBEDDINGS_REGISTRY: list[Type[LateInteractionTextEmbeddingBase]] = [Colbert, JinaColbert]
+
+    @classmethod
+    def list_supported_models(cls) -> list[dict[str, Any]]:
+        """
+        Lists the supported models.
+
+        Returns:
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
+
+            Example:
+                ```
+                [
+                    {
+                        "model": "colbert-ir/colbertv2.0",
+                        "dim": 128,
+                        "description": "Late interaction model",
+                        "license": "mit",
+                        "size_in_GB": 0.44,
+                        "sources": {
+                            "hf": "colbert-ir/colbertv2.0",
+                        },
+                        "model_file": "model.onnx",
+                    },
+                ]
+                ```
+        """
+        return [asdict(model) for model in cls._list_supported_models()]
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        result: list[DenseModelDescription] = []
+        for embedding in cls.EMBEDDINGS_REGISTRY:
+            result.extend(embedding._list_supported_models())
+        return result
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        **kwargs: Any,
+    ):
+        super().__init__(model_name, cache_dir, threads, **kwargs)
+        for EMBEDDING_MODEL_TYPE in self.EMBEDDINGS_REGISTRY:
+            supported_models = EMBEDDING_MODEL_TYPE._list_supported_models()
+            if any(model_name.lower() == model.model.lower() for model in supported_models):
+                self.model = EMBEDDING_MODEL_TYPE(
+                    model_name,
+                    cache_dir,
+                    threads=threads,
+                    providers=providers,
+                    cuda=cuda,
+                    device_ids=device_ids,
+                    lazy_load=lazy_load,
+                    **kwargs,
+                )
+                return
+
+        raise ValueError(
+            f"Model {model_name} is not supported in LateInteractionTextEmbedding."
+            "Please check the supported models using `LateInteractionTextEmbedding.list_supported_models()`"
+        )
+
+    def embed(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        """
+        Encode a list of documents into list of embeddings.
+        We use mean pooling with attention so that the model can handle variable-length inputs.
+
+        Args:
+            documents: Iterator of documents or single document to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per document
+        """
+        yield from self.model.embed(documents, batch_size, parallel, **kwargs)
+
+    def query_embed(self, query: Union[str, Iterable[str]], **kwargs: Any) -> Iterable[NumpyArray]:
+        """
+        Embeds queries
+
+        Args:
+            query (Union[str, Iterable[str]]): The query to embed, or an iterable e.g. list of queries.
+
+        Returns:
+            Iterable[NdArray]: The embeddings.
+        """
+
+        # This is model-specific, so that different models can have specialized implementations
+        yield from self.model.query_embed(query, **kwargs)
@@ -0,0 +1,5 @@
+from fastembed.late_interaction_multimodal.late_interaction_multimodal_embedding import (
+    LateInteractionMultimodalEmbedding,
+)
+
+__all__ = ["LateInteractionMultimodalEmbedding"]
@@ -0,0 +1,300 @@
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+
+import numpy as np
+from tokenizers import Encoding
+
+from fastembed.common import OnnxProvider, ImageInput
+from fastembed.common.onnx_model import OnnxOutputContext
+from fastembed.common.types import NumpyArray
+from fastembed.common.utils import define_cache_dir
+from fastembed.late_interaction_multimodal.late_interaction_multimodal_embedding_base import (
+    LateInteractionMultimodalEmbeddingBase,
+)
+from fastembed.late_interaction_multimodal.onnx_multimodal_model import (
+    OnnxMultimodalModel,
+    TextEmbeddingWorker,
+    ImageEmbeddingWorker,
+)
+from fastembed.common.model_description import DenseModelDescription, ModelSource
+
+supported_colpali_models: list[DenseModelDescription] = [
+    DenseModelDescription(
+        model="Qdrant/colpali-v1.3-fp16",
+        dim=128,
+        description="Text embeddings, Multimodal (text&image), English, 50 tokens query length truncation, 2024.",
+        license="mit",
+        size_in_GB=6.5,
+        sources=ModelSource(hf="Qdrant/colpali-v1.3-fp16"),
+        additional_files=["model.onnx_data"],
+        model_file="model.onnx",
+    ),
+]
+
+
+class ColPali(LateInteractionMultimodalEmbeddingBase, OnnxMultimodalModel[NumpyArray]):
+    QUERY_PREFIX = "Query: "
+    BOS_TOKEN = "<s>"
+    PAD_TOKEN = "<pad>"
+    QUERY_MARKER_TOKEN_ID = [2, 5098]
+    IMAGE_PLACEHOLDER_SIZE = (3, 448, 448)
+    EMPTY_TEXT_PLACEHOLDER = np.array(
+        [257152] * 1024 + [2, 50721, 573, 2416, 235265, 108]
+    )  # This is a tokenization of '<image>' * 1024 + '<bos>Describe the image.\n' line which is used as placeholder
+    # while processing an image
+    EVEN_ATTENTION_MASK = np.array([1] * 1030)
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        device_id: Optional[int] = None,
+        specific_model_path: Optional[str] = None,
+        **kwargs: Any,
+    ):
+        """
+        Args:
+            model_name (str): The name of the model to use.
+            cache_dir (str, optional): The path to the cache directory.
+                                       Can be set using the `FASTEMBED_CACHE_PATH` env variable.
+                                       Defaults to `fastembed_cache` in the system's temp directory.
+            threads (int, optional): The number of threads single onnxruntime session can use. Defaults to None.
+            providers (Optional[Sequence[OnnxProvider]], optional): The list of onnxruntime providers to use.
+                Mutually exclusive with the `cuda` and `device_ids` arguments. Defaults to None.
+            cuda (bool, optional): Whether to use cuda for inference. Mutually exclusive with `providers`
+                Defaults to False.
+            device_ids (Optional[list[int]], optional): The list of device ids to use for data parallel processing in
+                workers. Should be used with `cuda=True`, mutually exclusive with `providers`. Defaults to None.
+            lazy_load (bool, optional): Whether to load the model during class initialization or on demand.
+                Should be set to True when using multiple-gpu and parallel encoding. Defaults to False.
+            device_id (Optional[int], optional): The device id to use for loading the model in the worker process.
+
+        Raises:
+            ValueError: If the model_name is not in the format <org>/<model> e.g. BAAI/bge-base-en.
+        """
+
+        super().__init__(model_name, cache_dir, threads, **kwargs)
+        self.providers = providers
+        self.lazy_load = lazy_load
+
+        # List of device ids, that can be used for data parallel processing in workers
+        self.device_ids = device_ids
+        self.cuda = cuda
+
+        # This device_id will be used if we need to load model in current process
+        self.device_id: Optional[int] = None
+        if device_id is not None:
+            self.device_id = device_id
+        elif self.device_ids is not None:
+            self.device_id = self.device_ids[0]
+
+        self.model_description = self._get_model_description(model_name)
+        self.cache_dir = str(define_cache_dir(cache_dir))
+
+        self._model_dir = self.download_model(
+            self.model_description,
+            self.cache_dir,
+            local_files_only=self._local_files_only,
+            specific_model_path=specific_model_path,
+        )
+        self.mask_token_id = None
+        self.pad_token_id = None
+
+        if not self.lazy_load:
+            self.load_onnx_model()
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        """Lists the supported models.
+
+        Returns:
+            list[DenseModelDescription]: A list of DenseModelDescription objects containing the model information.
+        """
+        return supported_colpali_models
+
+    def load_onnx_model(self) -> None:
+        self._load_onnx_model(
+            model_dir=self._model_dir,
+            model_file=self.model_description.model_file,
+            threads=self.threads,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_id=self.device_id,
+        )
+
+    def _post_process_onnx_image_output(
+        self,
+        output: OnnxOutputContext,
+    ) -> Iterable[NumpyArray]:
+        """
+        Post-process the ONNX model output to convert it into a usable format.
+
+        Args:
+            output (OnnxOutputContext): The raw output from the ONNX model.
+
+        Returns:
+            Iterable[NumpyArray]: Post-processed output as NumPy arrays.
+        """
+        assert self.model_description.dim is not None, "Model dim is not defined"
+        return output.model_output.reshape(
+            output.model_output.shape[0], -1, self.model_description.dim
+        )
+
+    def _post_process_onnx_text_output(
+        self,
+        output: OnnxOutputContext,
+    ) -> Iterable[NumpyArray]:
+        """
+        Post-process the ONNX model output to convert it into a usable format.
+
+        Args:
+            output (OnnxOutputContext): The raw output from the ONNX model.
+
+        Returns:
+            Iterable[NumpyArray]: Post-processed output as NumPy arrays.
+        """
+        return output.model_output
+
+    def tokenize(self, documents: list[str], **kwargs: Any) -> list[Encoding]:
+        texts_query: list[str] = []
+        for query in documents:
+            query = self.BOS_TOKEN + self.QUERY_PREFIX + query + self.PAD_TOKEN * 10
+            query += "\n"
+
+            texts_query.append(query)
+        encoded = self.tokenizer.encode_batch(texts_query)  # type: ignore[union-attr]
+        return encoded
+
+    def _preprocess_onnx_text_input(
+        self, onnx_input: dict[str, NumpyArray], **kwargs: Any
+    ) -> dict[str, NumpyArray]:
+        onnx_input["input_ids"] = np.array(
+            [
+                self.QUERY_MARKER_TOKEN_ID + input_ids[2:].tolist()
+                for input_ids in onnx_input["input_ids"]
+            ]
+        )
+        empty_image_placeholder: NumpyArray = np.zeros(
+            self.IMAGE_PLACEHOLDER_SIZE, dtype=np.float32
+        )
+        onnx_input["pixel_values"] = np.array(
+            [empty_image_placeholder for _ in onnx_input["input_ids"]],
+        )
+        return onnx_input
+
+    def _preprocess_onnx_image_input(
+        self, onnx_input: dict[str, np.ndarray], **kwargs: Any
+    ) -> dict[str, NumpyArray]:
+        """
+        Add placeholders for text input when processing image data for ONNX.
+        Args:
+            onnx_input (Dict[str, NumpyArray]): Preprocessed image inputs.
+            **kwargs: Additional arguments.
+        Returns:
+            Dict[str, NumpyArray]: ONNX input with text placeholders.
+        """
+        onnx_input["input_ids"] = np.array(
+            [self.EMPTY_TEXT_PLACEHOLDER for _ in onnx_input["pixel_values"]]
+        )
+        onnx_input["attention_mask"] = np.array(
+            [self.EVEN_ATTENTION_MASK for _ in onnx_input["pixel_values"]]
+        )
+        return onnx_input
+
+    def embed_text(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        """
+        Encode a list of documents into list of embeddings.
+
+        Args:
+            documents: Iterator of documents or single document to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per document
+        """
+        yield from self._embed_documents(
+            model_name=self.model_name,
+            cache_dir=str(self.cache_dir),
+            documents=documents,
+            batch_size=batch_size,
+            parallel=parallel,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_ids=self.device_ids,
+            **kwargs,
+        )
+
+    def embed_image(
+        self,
+        images: Union[ImageInput, Iterable[ImageInput]],
+        batch_size: int = 16,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        """
+        Encode a list of images into list of embeddings.
+
+        Args:
+            images: Iterator of image paths or single image path to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per document
+        """
+        yield from self._embed_images(
+            model_name=self.model_name,
+            cache_dir=str(self.cache_dir),
+            images=images,
+            batch_size=batch_size,
+            parallel=parallel,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_ids=self.device_ids,
+            **kwargs,
+        )
+
+    @classmethod
+    def _get_text_worker_class(cls) -> Type[TextEmbeddingWorker[NumpyArray]]:
+        return ColPaliTextEmbeddingWorker
+
+    @classmethod
+    def _get_image_worker_class(cls) -> Type[ImageEmbeddingWorker[NumpyArray]]:
+        return ColPaliImageEmbeddingWorker
+
+
+class ColPaliTextEmbeddingWorker(TextEmbeddingWorker[NumpyArray]):
+    def init_embedding(self, model_name: str, cache_dir: str, **kwargs: Any) -> ColPali:
+        return ColPali(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )
+
+
+class ColPaliImageEmbeddingWorker(ImageEmbeddingWorker[NumpyArray]):
+    def init_embedding(self, model_name: str, cache_dir: str, **kwargs: Any) -> ColPali:
+        return ColPali(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )
@@ -0,0 +1,130 @@
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+from dataclasses import asdict
+
+from fastembed.common import OnnxProvider, ImageInput
+from fastembed.common.types import NumpyArray
+from fastembed.late_interaction_multimodal.colpali import ColPali
+
+from fastembed.late_interaction_multimodal.late_interaction_multimodal_embedding_base import (
+    LateInteractionMultimodalEmbeddingBase,
+)
+from fastembed.common.model_description import DenseModelDescription
+
+
+class LateInteractionMultimodalEmbedding(LateInteractionMultimodalEmbeddingBase):
+    EMBEDDINGS_REGISTRY: list[Type[LateInteractionMultimodalEmbeddingBase]] = [ColPali]
+
+    @classmethod
+    def list_supported_models(cls) -> list[dict[str, Any]]:
+        """
+        Lists the supported models.
+
+        Returns:
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
+
+            Example:
+                ```
+                [
+                    {
+                         "model": "Qdrant/colpali-v1.3-fp16",
+                         "dim": 128,
+                         "description": "Text embeddings, Unimodal (text), Aligned to image latent space, ColBERT-compatible, 512 tokens max, 2024.",
+                         "license": "mit",
+                         "size_in_GB": 6.06,
+                         "sources": {
+                            "hf": "Qdrant/colpali-v1.3-fp16",
+                         },
+                         "additional_files": [
+                            "model.onnx_data",
+                         ],
+                        "model_file": "model.onnx",
+                    },
+                ]
+                ```
+        """
+        return [asdict(model) for model in cls._list_supported_models()]
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        result: list[DenseModelDescription] = []
+        for embedding in cls.EMBEDDINGS_REGISTRY:
+            result.extend(embedding._list_supported_models())
+        return result
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        **kwargs: Any,
+    ):
+        super().__init__(model_name, cache_dir, threads, **kwargs)
+        for EMBEDDING_MODEL_TYPE in self.EMBEDDINGS_REGISTRY:
+            supported_models = EMBEDDING_MODEL_TYPE._list_supported_models()
+            if any(model_name.lower() == model.model.lower() for model in supported_models):
+                self.model = EMBEDDING_MODEL_TYPE(
+                    model_name,
+                    cache_dir,
+                    threads=threads,
+                    providers=providers,
+                    cuda=cuda,
+                    device_ids=device_ids,
+                    lazy_load=lazy_load,
+                    **kwargs,
+                )
+                return
+
+        raise ValueError(
+            f"Model {model_name} is not supported in LateInteractionMultimodalEmbedding."
+            "Please check the supported models using `LateInteractionMultimodalEmbedding.list_supported_models()`"
+        )
+
+    def embed_text(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        """
+        Encode a list of documents into list of embeddings.
+
+        Args:
+            documents: Iterator of documents or single document to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per document
+        """
+        yield from self.model.embed_text(documents, batch_size, parallel, **kwargs)
+
+    def embed_image(
+        self,
+        images: Union[ImageInput, Iterable[ImageInput]],
+        batch_size: int = 16,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        """
+        Encode a list of images into list of embeddings.
+
+        Args:
+            images: Iterator of image paths or single image path to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per image
+        """
+        yield from self.model.embed_image(images, batch_size, parallel, **kwargs)
@@ -0,0 +1,67 @@
+from typing import Iterable, Optional, Union, Any
+
+
+from fastembed.common import ImageInput
+from fastembed.common.model_description import DenseModelDescription
+from fastembed.common.model_management import ModelManagement
+from fastembed.common.types import NumpyArray
+
+
+class LateInteractionMultimodalEmbeddingBase(ModelManagement[DenseModelDescription]):
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        **kwargs: Any,
+    ):
+        self.model_name = model_name
+        self.cache_dir = cache_dir
+        self.threads = threads
+        self._local_files_only = kwargs.pop("local_files_only", False)
+
+    def embed_text(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        """
+        Embeds a list of documents into a list of embeddings.
+
+        Args:
+            documents (Iterable[str]): The list of texts to embed.
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+            **kwargs: Additional keyword argument to pass to the embed method.
+
+        Yields:
+            Iterable[NumpyArray]: The embeddings.
+        """
+        raise NotImplementedError()
+
+    def embed_image(
+        self,
+        images: Union[ImageInput, Iterable[ImageInput]],
+        batch_size: int = 16,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        """
+        Encode a list of images into list of embeddings.
+        Args:
+            images: Iterator of image paths or single image path to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per image
+        """
+        raise NotImplementedError()
@@ -0,0 +1,271 @@
+import contextlib
+import os
+from multiprocessing import get_all_start_methods
+from pathlib import Path
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+
+import numpy as np
+from PIL import Image
+from tokenizers import Encoding, Tokenizer
+
+from fastembed.common import OnnxProvider, ImageInput
+from fastembed.common.onnx_model import EmbeddingWorker, OnnxModel, OnnxOutputContext, T
+from fastembed.common.preprocessor_utils import load_tokenizer, load_preprocessor
+from fastembed.common.types import NumpyArray
+from fastembed.common.utils import iter_batch
+from fastembed.image.transform.operators import Compose
+from fastembed.parallel_processor import ParallelWorkerPool
+
+
+class OnnxMultimodalModel(OnnxModel[T]):
+    ONNX_OUTPUT_NAMES: Optional[list[str]] = None
+
+    def __init__(self) -> None:
+        super().__init__()
+        self.tokenizer: Optional[Tokenizer] = None
+        self.processor: Optional[Compose] = None
+        self.special_token_to_id: dict[str, int] = {}
+
+    def _preprocess_onnx_text_input(
+        self, onnx_input: dict[str, NumpyArray], **kwargs: Any
+    ) -> dict[str, NumpyArray]:
+        """
+        Preprocess the onnx input.
+        """
+        return onnx_input
+
+    def _preprocess_onnx_image_input(
+        self, onnx_input: dict[str, NumpyArray], **kwargs: Any
+    ) -> dict[str, NumpyArray]:
+        """
+        Preprocess the onnx input.
+        """
+        return onnx_input
+
+    @classmethod
+    def _get_text_worker_class(cls) -> Type["TextEmbeddingWorker[T]"]:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    @classmethod
+    def _get_image_worker_class(cls) -> Type["ImageEmbeddingWorker[T]"]:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def _post_process_onnx_image_output(self, output: OnnxOutputContext) -> Iterable[T]:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def _post_process_onnx_text_output(self, output: OnnxOutputContext) -> Iterable[T]:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def _load_onnx_model(
+        self,
+        model_dir: Path,
+        model_file: str,
+        threads: Optional[int],
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_id: Optional[int] = None,
+    ) -> None:
+        super()._load_onnx_model(
+            model_dir=model_dir,
+            model_file=model_file,
+            threads=threads,
+            providers=providers,
+            cuda=cuda,
+            device_id=device_id,
+        )
+        self.tokenizer, self.special_token_to_id = load_tokenizer(model_dir=model_dir)
+        assert self.tokenizer is not None
+        self.processor = load_preprocessor(model_dir=model_dir)
+
+    def load_onnx_model(self) -> None:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def tokenize(self, documents: list[str], **kwargs: Any) -> list[Encoding]:
+        return self.tokenizer.encode_batch(documents)  # type: ignore[union-attr]
+
+    def onnx_embed_text(
+        self,
+        documents: list[str],
+        **kwargs: Any,
+    ) -> OnnxOutputContext:
+        encoded = self.tokenize(documents, **kwargs)
+        input_ids = np.array([e.ids for e in encoded])
+        attention_mask = np.array([e.attention_mask for e in encoded])  # type: ignore[union-attr]
+        input_names = {node.name for node in self.model.get_inputs()}  # type: ignore[union-attr]
+        onnx_input: dict[str, NumpyArray] = {
+            "input_ids": np.array(input_ids, dtype=np.int64),
+        }
+        if "attention_mask" in input_names:
+            onnx_input["attention_mask"] = np.array(attention_mask, dtype=np.int64)
+        if "token_type_ids" in input_names:
+            onnx_input["token_type_ids"] = np.array(
+                [np.zeros(len(e), dtype=np.int64) for e in input_ids], dtype=np.int64
+            )
+
+        onnx_input = self._preprocess_onnx_text_input(onnx_input, **kwargs)
+        model_output = self.model.run(self.ONNX_OUTPUT_NAMES, onnx_input)  # type: ignore[union-attr]
+        return OnnxOutputContext(
+            model_output=model_output[0],
+            attention_mask=onnx_input.get("attention_mask", attention_mask),
+            input_ids=onnx_input.get("input_ids", input_ids),
+        )
+
+    def _embed_documents(
+        self,
+        model_name: str,
+        cache_dir: str,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        **kwargs: Any,
+    ) -> Iterable[T]:
+        is_small = False
+
+        if isinstance(documents, str):
+            documents = [documents]
+            is_small = True
+
+        if isinstance(documents, list):
+            if len(documents) < batch_size:
+                is_small = True
+
+        if parallel is None or is_small:
+            if not hasattr(self, "model") or self.model is None:
+                self.load_onnx_model()
+            for batch in iter_batch(documents, batch_size):
+                yield from self._post_process_onnx_text_output(self.onnx_embed_text(batch))
+        else:
+            if parallel == 0:
+                parallel = os.cpu_count()
+
+            start_method = "forkserver" if "forkserver" in get_all_start_methods() else "spawn"
+            params = {
+                "model_name": model_name,
+                "cache_dir": cache_dir,
+                "providers": providers,
+                **kwargs,
+            }
+
+            pool = ParallelWorkerPool(
+                num_workers=parallel or 1,
+                worker=self._get_text_worker_class(),
+                cuda=cuda,
+                device_ids=device_ids,
+                start_method=start_method,
+            )
+            for batch in pool.ordered_map(iter_batch(documents, batch_size), **params):
+                yield from self._post_process_onnx_text_output(batch)  # type: ignore
+
+    def onnx_embed_image(self, images: list[ImageInput], **kwargs: Any) -> OnnxOutputContext:
+        with contextlib.ExitStack():
+            image_files = [
+                Image.open(image) if not isinstance(image, Image.Image) else image
+                for image in images
+            ]
+            assert self.processor is not None, "Processor is not initialized"
+            encoded = np.array(self.processor(image_files))
+        onnx_input = {"pixel_values": encoded}
+        onnx_input = self._preprocess_onnx_image_input(onnx_input, **kwargs)
+        model_output = self.model.run(None, onnx_input)  # type: ignore[union-attr]
+        embeddings = model_output[0].reshape(len(images), -1)
+        return OnnxOutputContext(model_output=embeddings)
+
+    def _embed_images(
+        self,
+        model_name: str,
+        cache_dir: str,
+        images: Union[Iterable[ImageInput], ImageInput],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        **kwargs: Any,
+    ) -> Iterable[T]:
+        is_small = False
+
+        if isinstance(images, (str, Path, Image.Image)):
+            images = [images]
+            is_small = True
+
+        if isinstance(images, list) and len(images) < batch_size:
+            is_small = True
+
+        if parallel is None or is_small:
+            if not hasattr(self, "model") or self.model is None:
+                self.load_onnx_model()
+
+            for batch in iter_batch(images, batch_size):
+                yield from self._post_process_onnx_image_output(self.onnx_embed_image(batch))
+        else:
+            if parallel == 0:
+                parallel = os.cpu_count()
+
+            start_method = "forkserver" if "forkserver" in get_all_start_methods() else "spawn"
+            params = {
+                "model_name": model_name,
+                "cache_dir": cache_dir,
+                "providers": providers,
+                **kwargs,
+            }
+
+            pool = ParallelWorkerPool(
+                num_workers=parallel or 1,
+                worker=self._get_image_worker_class(),
+                cuda=cuda,
+                device_ids=device_ids,
+                start_method=start_method,
+            )
+            for batch in pool.ordered_map(iter_batch(images, batch_size), **params):
+                yield from self._post_process_onnx_image_output(batch)  # type: ignore
+
+
+class TextEmbeddingWorker(EmbeddingWorker[T]):
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ):
+        self.model: OnnxMultimodalModel
+        super().__init__(model_name, cache_dir, **kwargs)
+
+    def init_embedding(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ) -> OnnxMultimodalModel:
+        raise NotImplementedError()
+
+    def process(self, items: Iterable[tuple[int, Any]]) -> Iterable[tuple[int, Any]]:
+        for idx, batch in items:
+            onnx_output = self.model.onnx_embed_text(batch)
+            yield idx, onnx_output
+
+
+class ImageEmbeddingWorker(EmbeddingWorker[T]):
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ):
+        self.model: OnnxMultimodalModel
+        super().__init__(model_name, cache_dir, **kwargs)
+
+    def init_embedding(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ) -> OnnxMultimodalModel:
+        raise NotImplementedError()
+
+    def process(self, items: Iterable[tuple[int, Any]]) -> Iterable[tuple[int, Any]]:
+        for idx, batch in items:
+            embeddings = self.model.onnx_embed_image(batch)
+            yield idx, embeddings
@@ -0,0 +1,252 @@
+import logging
+import os
+from collections import defaultdict
+from copy import deepcopy
+from enum import Enum
+from multiprocessing import Queue, get_context
+from multiprocessing.context import BaseContext
+from multiprocessing.process import BaseProcess
+from multiprocessing.sharedctypes import Synchronized as BaseValue
+from queue import Empty
+from typing import Any, Iterable, Optional, Type
+
+
+# Single item should be processed in less than:
+processing_timeout = 10 * 60  # seconds
+
+max_internal_batch_size = 200
+
+
+class QueueSignals(str, Enum):
+    stop = "stop"
+    confirm = "confirm"
+    error = "error"
+
+
+class Worker:
+    @classmethod
+    def start(cls, *args: Any, **kwargs: Any) -> "Worker":
+        raise NotImplementedError()
+
+    def process(self, items: Iterable[tuple[int, Any]]) -> Iterable[tuple[int, Any]]:
+        raise NotImplementedError()
+
+
+def _worker(
+    worker_class: Type[Worker],
+    input_queue: Queue,
+    output_queue: Queue,
+    num_active_workers: BaseValue,
+    worker_id: int,
+    kwargs: Optional[dict[str, Any]] = None,
+) -> None:
+    """
+    A worker that pulls data pints off the input queue, and places the execution result on the output queue.
+    When there are no data pints left on the input queue, it decrements
+    num_active_workers to signal completion.
+    """
+
+    if kwargs is None:
+        kwargs = {}
+
+    logging.info(
+        f"Reader worker: {worker_id} PID: {os.getpid()} Device: {kwargs.get('device_id', 'CPU')}"
+    )
+    try:
+        worker = worker_class.start(**kwargs)
+
+        # Keep going until you get an item that's None.
+        def input_queue_iterable() -> Iterable[Any]:
+            while True:
+                item = input_queue.get()
+                if item == QueueSignals.stop:
+                    break
+                yield item
+
+        for processed_item in worker.process(input_queue_iterable()):
+            output_queue.put(processed_item)
+    except Exception as e:  # pylint: disable=broad-except
+        logging.exception(e)
+        output_queue.put(QueueSignals.error)
+    finally:
+        # It's important that we close and join the queue here before
+        # decrementing num_active_workers. Otherwise our parent may join us
+        # before the queue's feeder thread has passed all buffered items to
+        # the underlying pipe resulting in a deadlock.
+        #
+        # See:
+        # https://docs.python.org/3.6/library/multiprocessing.html?highlight=process#pipes-and-queues
+        # https://docs.python.org/3.6/library/multiprocessing.html?highlight=process#programming-guidelines
+        input_queue.close()
+        output_queue.close()
+        input_queue.join_thread()
+        output_queue.join_thread()
+
+        with num_active_workers.get_lock():
+            num_active_workers.value -= 1
+
+        logging.info(f"Reader worker {worker_id} finished")
+
+
+class ParallelWorkerPool:
+    def __init__(
+        self,
+        num_workers: int,
+        worker: Type[Worker],
+        start_method: Optional[str] = None,
+        device_ids: Optional[list[int]] = None,
+        cuda: bool = False,
+    ):
+        self.worker_class = worker
+        self.num_workers = num_workers
+        self.input_queue: Optional[Queue] = None
+        self.output_queue: Optional[Queue] = None
+        self.ctx: BaseContext = get_context(start_method)
+        self.processes: list[BaseProcess] = []
+        self.queue_size = self.num_workers * max_internal_batch_size
+        self.emergency_shutdown = False
+        self.device_ids = device_ids
+        self.cuda = cuda
+        self.num_active_workers: Optional[BaseValue] = None
+
+    def start(self, **kwargs: Any) -> None:
+        self.input_queue = self.ctx.Queue(self.queue_size)
+        self.output_queue = self.ctx.Queue(self.queue_size)
+
+        ctx_value = self.ctx.Value("i", self.num_workers)
+        assert isinstance(ctx_value, BaseValue)
+        self.num_active_workers = ctx_value
+
+        for worker_id in range(0, self.num_workers):
+            worker_kwargs = deepcopy(kwargs)
+            if self.device_ids:
+                device_id = self.device_ids[worker_id % len(self.device_ids)]
+                worker_kwargs["device_id"] = device_id
+                worker_kwargs["cuda"] = self.cuda
+
+            assert hasattr(self.ctx, "Process")
+            process = self.ctx.Process(
+                target=_worker,
+                args=(
+                    self.worker_class,
+                    self.input_queue,
+                    self.output_queue,
+                    self.num_active_workers,
+                    worker_id,
+                    worker_kwargs,
+                ),
+            )
+            process.start()
+            self.processes.append(process)
+
+    def ordered_map(self, stream: Iterable[Any], *args: Any, **kwargs: Any) -> Iterable[Any]:
+        buffer: defaultdict[int, Any] = defaultdict(Any)  # type: ignore
+        next_expected = 0
+
+        for idx, item in self.semi_ordered_map(stream, *args, **kwargs):
+            buffer[idx] = item
+            while next_expected in buffer:
+                yield buffer.pop(next_expected)
+                next_expected += 1
+
+    def semi_ordered_map(
+        self, stream: Iterable[Any], *args: Any, **kwargs: Any
+    ) -> Iterable[tuple[int, Any]]:
+        try:
+            self.start(**kwargs)
+
+            assert self.input_queue is not None, "Input queue was not initialized"
+            assert self.output_queue is not None, "Output queue was not initialized"
+
+            pushed = 0
+            read = 0
+            for idx, item in enumerate(stream):
+                self.check_worker_health()
+                if pushed - read < self.queue_size:
+                    try:
+                        out_item = self.output_queue.get_nowait()
+                    except Empty:
+                        out_item = None
+                else:
+                    try:
+                        out_item = self.output_queue.get(timeout=processing_timeout)
+                    except Empty as e:
+                        self.join_or_terminate()
+                        raise e
+
+                if out_item is not None:
+                    if out_item == QueueSignals.error:
+                        self.join_or_terminate()
+                        raise RuntimeError("Thread unexpectedly terminated")
+                    yield out_item
+                    read += 1
+
+                self.input_queue.put((idx, item))
+                pushed += 1
+
+            for _ in range(self.num_workers):
+                self.input_queue.put(QueueSignals.stop)
+
+            while read < pushed:
+                self.check_worker_health()
+                out_item = self.output_queue.get(timeout=processing_timeout)
+                if out_item == QueueSignals.error:
+                    self.join_or_terminate()
+                    raise RuntimeError("Thread unexpectedly terminated")
+                yield out_item
+                read += 1
+        finally:
+            assert self.input_queue is not None, "Input queue is None"
+            assert self.output_queue is not None, "Output queue is None"
+            self.join()
+            self.input_queue.close()
+            self.output_queue.close()
+            if self.emergency_shutdown:
+                self.input_queue.cancel_join_thread()
+                self.output_queue.cancel_join_thread()
+            else:
+                self.input_queue.join_thread()
+                self.output_queue.join_thread()
+
+    def check_worker_health(self) -> None:
+        """
+        Checks if any worker process has terminated unexpectedly
+        """
+        for process in self.processes:
+            if not process.is_alive() and process.exitcode != 0:
+                self.emergency_shutdown = True
+                self.join_or_terminate()
+                raise RuntimeError(
+                    f"Worker PID: {process.pid} terminated unexpectedly with code {process.exitcode}"
+                )
+
+    def join_or_terminate(self, timeout: Optional[int] = 1) -> None:
+        """
+        Emergency shutdown
+        @param timeout:
+        @return:
+        """
+        for process in self.processes:
+            process.join(timeout=timeout)
+            if process.is_alive():
+                process.terminate()
+        self.processes.clear()
+
+    def join(self) -> None:
+        for process in self.processes:
+            process.join()
+        self.processes.clear()
+
+    def __del__(self) -> None:
+        """
+        Terminate processes if the user hasn't joined. This is necessary as
+        leaving stray processes running can corrupt shared state. In brief,
+        we've observed shared memory counters being reused (when the memory was
+        free from the perspective of the parent process) while the stray
+        workers still held a reference to them.
+        For a discussion of using destructors in Python in this manner, see
+        https://eli.thegreenplace.net/2009/06/12/safely-using-destructors-in-python/.
+        """
+        for process in self.processes:
+            if process.is_alive():
+                process.terminate()
@@ -0,0 +1 @@
+partial
@@ -0,0 +1,3 @@
+from fastembed.rerank.cross_encoder.text_cross_encoder import TextCrossEncoder
+
+__all__ = ["TextCrossEncoder"]
@@ -0,0 +1,46 @@
+from typing import Optional, Sequence, Any
+
+from fastembed.common import OnnxProvider
+from fastembed.common.model_description import BaseModelDescription
+from fastembed.rerank.cross_encoder.onnx_text_cross_encoder import OnnxTextCrossEncoder
+
+
+class CustomTextCrossEncoder(OnnxTextCrossEncoder):
+    SUPPORTED_MODELS: list[BaseModelDescription] = []
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        device_id: Optional[int] = None,
+        specific_model_path: Optional[str] = None,
+        **kwargs: Any,
+    ):
+        super().__init__(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=threads,
+            providers=providers,
+            cuda=cuda,
+            device_ids=device_ids,
+            lazy_load=lazy_load,
+            device_id=device_id,
+            specific_model_path=specific_model_path,
+            **kwargs,
+        )
+
+    @classmethod
+    def _list_supported_models(cls) -> list[BaseModelDescription]:
+        return cls.SUPPORTED_MODELS
+
+    @classmethod
+    def add_model(
+        cls,
+        model_description: BaseModelDescription,
+    ) -> None:
+        cls.SUPPORTED_MODELS.append(model_description)
@@ -0,0 +1,215 @@
+from typing import Any, Iterable, Optional, Sequence, Type
+
+from loguru import logger
+
+from fastembed.common import OnnxProvider
+from fastembed.common.onnx_model import OnnxOutputContext
+from fastembed.common.utils import define_cache_dir
+from fastembed.rerank.cross_encoder.onnx_text_model import (
+    OnnxCrossEncoderModel,
+    TextRerankerWorker,
+)
+from fastembed.rerank.cross_encoder.text_cross_encoder_base import TextCrossEncoderBase
+from fastembed.common.model_description import BaseModelDescription, ModelSource
+
+supported_onnx_models: list[BaseModelDescription] = [
+    BaseModelDescription(
+        model="Xenova/ms-marco-MiniLM-L-6-v2",
+        description="MiniLM-L-6-v2 model optimized for re-ranking tasks.",
+        license="apache-2.0",
+        size_in_GB=0.08,
+        sources=ModelSource(hf="Xenova/ms-marco-MiniLM-L-6-v2"),
+        model_file="onnx/model.onnx",
+    ),
+    BaseModelDescription(
+        model="Xenova/ms-marco-MiniLM-L-12-v2",
+        description="MiniLM-L-12-v2 model optimized for re-ranking tasks.",
+        license="apache-2.0",
+        size_in_GB=0.12,
+        sources=ModelSource(hf="Xenova/ms-marco-MiniLM-L-12-v2"),
+        model_file="onnx/model.onnx",
+    ),
+    BaseModelDescription(
+        model="BAAI/bge-reranker-base",
+        description="BGE reranker base model for cross-encoder re-ranking.",
+        license="mit",
+        size_in_GB=1.04,
+        sources=ModelSource(hf="BAAI/bge-reranker-base"),
+        model_file="onnx/model.onnx",
+    ),
+    BaseModelDescription(
+        model="jinaai/jina-reranker-v1-tiny-en",
+        description="Designed for blazing-fast re-ranking with 8K context length and fewer parameters than jina-reranker-v1-turbo-en.",
+        license="apache-2.0",
+        size_in_GB=0.13,
+        sources=ModelSource(hf="jinaai/jina-reranker-v1-tiny-en"),
+        model_file="onnx/model.onnx",
+    ),
+    BaseModelDescription(
+        model="jinaai/jina-reranker-v1-turbo-en",
+        description="Designed for blazing-fast re-ranking with 8K context length.",
+        license="apache-2.0",
+        size_in_GB=0.15,
+        sources=ModelSource(hf="jinaai/jina-reranker-v1-turbo-en"),
+        model_file="onnx/model.onnx",
+    ),
+    BaseModelDescription(
+        model="jinaai/jina-reranker-v2-base-multilingual",
+        description="A multi-lingual reranker model for cross-encoder re-ranking with 1K context length and sliding window",
+        license="cc-by-nc-4.0",
+        size_in_GB=1.11,
+        sources=ModelSource(hf="jinaai/jina-reranker-v2-base-multilingual"),
+        model_file="onnx/model.onnx",
+    ),
+]
+
+
+class OnnxTextCrossEncoder(TextCrossEncoderBase, OnnxCrossEncoderModel):
+    @classmethod
+    def _list_supported_models(cls) -> list[BaseModelDescription]:
+        """Lists the supported models.
+
+        Returns:
+            list[BaseModelDescription]: A list of BaseModelDescription objects containing the model information.
+        """
+        return supported_onnx_models
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        device_id: Optional[int] = None,
+        specific_model_path: Optional[str] = None,
+        **kwargs: Any,
+    ):
+        """
+        Args:
+            model_name (str): The name of the model to use.
+            cache_dir (str, optional): The path to the cache directory.
+                                       Can be set using the `FASTEMBED_CACHE_PATH` env variable.
+                                       Defaults to `fastembed_cache` in the system's temp directory.
+            threads (int, optional): The number of threads single onnxruntime session can use. Defaults to None.
+            providers (Optional[Sequence[OnnxProvider]], optional): The list of onnxruntime providers to use.
+                Mutually exclusive with the `cuda` and `device_ids` arguments. Defaults to None.
+            cuda (bool, optional): Whether to use cuda for inference. Mutually exclusive with `providers`
+                Defaults to False.
+            device_ids (Optional[list[int]], optional): The list of device ids to use for data parallel processing in
+                workers. Should be used with `cuda=True`, mutually exclusive with `providers`. Defaults to None.
+            lazy_load (bool, optional): Whether to load the model during class initialization or on demand.
+                Should be set to True when using multiple-gpu and parallel encoding. Defaults to False.
+            device_id (Optional[int], optional): The device id to use for loading the model in the worker process.
+            specific_model_path (Optional[str], optional): The specific path to the onnx model dir if it should be imported from somewhere else
+
+        Raises:
+            ValueError: If the model_name is not in the format <org>/<model> e.g. Xenova/ms-marco-MiniLM-L-6-v2.
+        """
+        super().__init__(model_name, cache_dir, threads, **kwargs)
+        self.providers = providers
+        self.lazy_load = lazy_load
+
+        # List of device ids, that can be used for data parallel processing in workers
+        self.device_ids = device_ids
+        self.cuda = cuda
+
+        if self.device_ids is not None and len(self.device_ids) > 1:
+            logger.warning(
+                "Parallel execution is currently not supported for cross encoders, "
+                f"only the first device will be used for inference: {self.device_ids[0]}."
+            )
+
+        # This device_id will be used if we need to load model in current process
+        self.device_id: Optional[int] = None
+        if device_id is not None:
+            self.device_id = device_id
+        elif self.device_ids is not None:
+            self.device_id = self.device_ids[0]
+
+        self.model_description = self._get_model_description(model_name)
+        self.cache_dir = str(define_cache_dir(cache_dir))
+        self._model_dir = self.download_model(
+            self.model_description,
+            self.cache_dir,
+            local_files_only=self._local_files_only,
+            specific_model_path=specific_model_path,
+        )
+
+        if not self.lazy_load:
+            self.load_onnx_model()
+
+    def load_onnx_model(self) -> None:
+        self._load_onnx_model(
+            model_dir=self._model_dir,
+            model_file=self.model_description.model_file,
+            threads=self.threads,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_id=self.device_id,
+        )
+
+    def rerank(
+        self,
+        query: str,
+        documents: Iterable[str],
+        batch_size: int = 64,
+        **kwargs: Any,
+    ) -> Iterable[float]:
+        """Reranks documents based on their relevance to a given query.
+
+        Args:
+            query (str): The query string to which document relevance is calculated.
+            documents (Iterable[str]): Iterable of documents to be reranked.
+            batch_size (int, optional): The number of documents processed in each batch. Higher batch sizes improve speed
+                                        but require more memory. Default is 64.
+        Returns:
+            Iterable[float]: An iterable of relevance scores for each document.
+        """
+
+        yield from self._rerank_documents(
+            query=query, documents=documents, batch_size=batch_size, **kwargs
+        )
+
+    def rerank_pairs(
+        self,
+        pairs: Iterable[tuple[str, str]],
+        batch_size: int = 64,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[float]:
+        yield from self._rerank_pairs(
+            model_name=self.model_name,
+            cache_dir=str(self.cache_dir),
+            pairs=pairs,
+            batch_size=batch_size,
+            parallel=parallel,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_ids=self.device_ids,
+            **kwargs,
+        )
+
+    @classmethod
+    def _get_worker_class(cls) -> Type[TextRerankerWorker]:
+        return TextCrossEncoderWorker
+
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[float]:
+        return (float(elem) for elem in output.model_output)
+
+
+class TextCrossEncoderWorker(TextRerankerWorker):
+    def init_embedding(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ) -> OnnxTextCrossEncoder:
+        return OnnxTextCrossEncoder(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )
@@ -0,0 +1,169 @@
+import os
+from multiprocessing import get_all_start_methods
+from pathlib import Path
+from typing import Any, Iterable, Optional, Sequence, Type
+
+import numpy as np
+from tokenizers import Encoding
+
+from fastembed.common.onnx_model import (
+    EmbeddingWorker,
+    OnnxModel,
+    OnnxOutputContext,
+    OnnxProvider,
+)
+from fastembed.common.types import NumpyArray
+from fastembed.common.preprocessor_utils import load_tokenizer
+from fastembed.common.utils import iter_batch
+from fastembed.parallel_processor import ParallelWorkerPool
+
+
+class OnnxCrossEncoderModel(OnnxModel[float]):
+    ONNX_OUTPUT_NAMES: Optional[list[str]] = None
+
+    @classmethod
+    def _get_worker_class(cls) -> Type["TextRerankerWorker"]:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def _load_onnx_model(
+        self,
+        model_dir: Path,
+        model_file: str,
+        threads: Optional[int],
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_id: Optional[int] = None,
+    ) -> None:
+        super()._load_onnx_model(
+            model_dir=model_dir,
+            model_file=model_file,
+            threads=threads,
+            providers=providers,
+            cuda=cuda,
+            device_id=device_id,
+        )
+        self.tokenizer, _ = load_tokenizer(model_dir=model_dir)
+        assert self.tokenizer is not None
+
+    def tokenize(self, pairs: list[tuple[str, str]], **_: Any) -> list[Encoding]:
+        return self.tokenizer.encode_batch(pairs)  # type: ignore[union-attr]
+
+    def _build_onnx_input(self, tokenized_input: list[Encoding]) -> dict[str, NumpyArray]:
+        input_names: set[str] = {node.name for node in self.model.get_inputs()}  # type: ignore[union-attr]
+        inputs: dict[str, NumpyArray] = {
+            "input_ids": np.array([enc.ids for enc in tokenized_input], dtype=np.int64),
+        }
+        if "token_type_ids" in input_names:
+            inputs["token_type_ids"] = np.array(
+                [enc.type_ids for enc in tokenized_input], dtype=np.int64
+            )
+        if "attention_mask" in input_names:
+            inputs["attention_mask"] = np.array(
+                [enc.attention_mask for enc in tokenized_input], dtype=np.int64
+            )
+        return inputs
+
+    def onnx_embed(self, query: str, documents: list[str], **kwargs: Any) -> OnnxOutputContext:
+        pairs = [(query, doc) for doc in documents]
+        return self.onnx_embed_pairs(pairs, **kwargs)
+
+    def onnx_embed_pairs(self, pairs: list[tuple[str, str]], **kwargs: Any) -> OnnxOutputContext:
+        tokenized_input = self.tokenize(pairs, **kwargs)
+        inputs = self._build_onnx_input(tokenized_input)
+        onnx_input = self._preprocess_onnx_input(inputs, **kwargs)
+        outputs = self.model.run(self.ONNX_OUTPUT_NAMES, onnx_input)  # type: ignore[union-attr]
+        relevant_output = outputs[0]
+        scores: NumpyArray = relevant_output[:, 0]
+        return OnnxOutputContext(model_output=scores)
+
+    def _rerank_documents(
+        self, query: str, documents: Iterable[str], batch_size: int, **kwargs: Any
+    ) -> Iterable[float]:
+        if not hasattr(self, "model") or self.model is None:
+            self.load_onnx_model()
+        for batch in iter_batch(documents, batch_size):
+            yield from self._post_process_onnx_output(self.onnx_embed(query, batch, **kwargs))
+
+    def _rerank_pairs(
+        self,
+        model_name: str,
+        cache_dir: str,
+        pairs: Iterable[tuple[str, str]],
+        batch_size: int,
+        parallel: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        **kwargs: Any,
+    ) -> Iterable[float]:
+        is_small = False
+
+        if isinstance(pairs, tuple):
+            pairs = [pairs]
+            is_small = True
+
+        if isinstance(pairs, list):
+            if len(pairs) < batch_size:
+                is_small = True
+
+        if parallel is None or is_small:
+            if not hasattr(self, "model") or self.model is None:
+                self.load_onnx_model()
+            for batch in iter_batch(pairs, batch_size):
+                yield from self._post_process_onnx_output(self.onnx_embed_pairs(batch, **kwargs))
+        else:
+            if parallel == 0:
+                parallel = os.cpu_count()
+
+            start_method = "forkserver" if "forkserver" in get_all_start_methods() else "spawn"
+            params = {
+                "model_name": model_name,
+                "cache_dir": cache_dir,
+                "providers": providers,
+                **kwargs,
+            }
+
+            pool = ParallelWorkerPool(
+                num_workers=parallel or 1,
+                worker=self._get_worker_class(),
+                cuda=cuda,
+                device_ids=device_ids,
+                start_method=start_method,
+            )
+            for batch in pool.ordered_map(iter_batch(pairs, batch_size), **params):
+                yield from self._post_process_onnx_output(batch)  # type: ignore
+
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[float]:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def _preprocess_onnx_input(
+        self, onnx_input: dict[str, NumpyArray], **kwargs: Any
+    ) -> dict[str, NumpyArray]:
+        """
+        Preprocess the onnx input.
+        """
+        return onnx_input
+
+
+class TextRerankerWorker(EmbeddingWorker[float]):
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ):
+        self.model: OnnxCrossEncoderModel
+        super().__init__(model_name, cache_dir, **kwargs)
+
+    def init_embedding(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ) -> OnnxCrossEncoderModel:
+        raise NotImplementedError()
+
+    def process(self, items: Iterable[tuple[int, Any]]) -> Iterable[tuple[int, Any]]:
+        for idx, batch in items:
+            onnx_output = self.model.onnx_embed_pairs(batch)
+            yield idx, onnx_output
@@ -0,0 +1,163 @@
+from typing import Any, Iterable, Optional, Sequence, Type
+from dataclasses import asdict
+
+from fastembed.common import OnnxProvider
+from fastembed.rerank.cross_encoder.onnx_text_cross_encoder import OnnxTextCrossEncoder
+from fastembed.rerank.cross_encoder.custom_text_cross_encoder import CustomTextCrossEncoder
+
+from fastembed.rerank.cross_encoder.text_cross_encoder_base import TextCrossEncoderBase
+from fastembed.common.model_description import (
+    ModelSource,
+    BaseModelDescription,
+)
+
+
+class TextCrossEncoder(TextCrossEncoderBase):
+    CROSS_ENCODER_REGISTRY: list[Type[TextCrossEncoderBase]] = [
+        OnnxTextCrossEncoder,
+        CustomTextCrossEncoder,
+    ]
+
+    @classmethod
+    def list_supported_models(cls) -> list[dict[str, Any]]:
+        """Lists the supported models.
+
+        Returns:
+            list[BaseModelDescription]: A list of dictionaries containing the model information.
+
+            Example:
+                ```
+                [
+                    {
+                        "model": "Xenova/ms-marco-MiniLM-L-6-v2",
+                        "size_in_GB": 0.08,
+                        "sources": {
+                            "hf": "Xenova/ms-marco-MiniLM-L-6-v2",
+                        },
+                        "model_file": "onnx/model.onnx",
+                        "description": "MiniLM-L-6-v2 model optimized for re-ranking tasks.",
+                        "license": "apache-2.0",
+                    }
+                ]
+                ```
+        """
+        return [asdict(model) for model in cls._list_supported_models()]
+
+    @classmethod
+    def _list_supported_models(cls) -> list[BaseModelDescription]:
+        result: list[BaseModelDescription] = []
+        for encoder in cls.CROSS_ENCODER_REGISTRY:
+            result.extend(encoder._list_supported_models())
+        return result
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        **kwargs: Any,
+    ):
+        super().__init__(model_name, cache_dir, threads, **kwargs)
+
+        for CROSS_ENCODER_TYPE in self.CROSS_ENCODER_REGISTRY:
+            supported_models = CROSS_ENCODER_TYPE._list_supported_models()
+            if any(model_name.lower() == model.model.lower() for model in supported_models):
+                self.model = CROSS_ENCODER_TYPE(
+                    model_name=model_name,
+                    cache_dir=cache_dir,
+                    threads=threads,
+                    providers=providers,
+                    cuda=cuda,
+                    device_ids=device_ids,
+                    lazy_load=lazy_load,
+                    **kwargs,
+                )
+                return
+
+        raise ValueError(
+            f"Model {model_name} is not supported in TextCrossEncoder."
+            "Please check the supported models using `TextCrossEncoder.list_supported_models()`"
+        )
+
+    def rerank(
+        self, query: str, documents: Iterable[str], batch_size: int = 64, **kwargs: Any
+    ) -> Iterable[float]:
+        """Rerank a list of documents based on a query.
+
+        Args:
+            query: Query to rerank the documents against
+            documents: Iterator of documents to rerank
+            batch_size: Batch size for reranking
+
+        Returns:
+            Iterable of scores for each document
+        """
+        yield from self.model.rerank(query, documents, batch_size=batch_size, **kwargs)
+
+    def rerank_pairs(
+        self,
+        pairs: Iterable[tuple[str, str]],
+        batch_size: int = 64,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[float]:
+        """
+        Rerank a list of query-document pairs.
+
+        Args:
+            pairs (Iterable[tuple[str, str]]): An iterable of tuples, where each tuple contains a query and a document
+                to be scored together.
+            batch_size (int, optional): The number of query-document pairs to process in a single batch. Defaults to 64.
+            parallel (Optional[int], optional): The number of parallel processes to use for reranking.
+                If None, parallelization is disabled. Defaults to None.
+            **kwargs (Any): Additional arguments to pass to the underlying reranking model.
+
+        Returns:
+            Iterable[float]: An iterable of scores corresponding to each query-document pair in the input.
+            Higher scores indicate a stronger match between the query and the document.
+
+        Example:
+            >>> encoder = TextCrossEncoder("Xenova/ms-marco-MiniLM-L-6-v2")
+            >>> pairs = [("What is AI?", "Artificial intelligence is ..."), ("What is ML?", "Machine learning is ...")]
+            >>> scores = list(encoder.rerank_pairs(pairs))
+            >>> print(list(map(lambda x: round(x, 2), scores)))
+            [-1.24, -10.6]
+        """
+        yield from self.model.rerank_pairs(
+            pairs, batch_size=batch_size, parallel=parallel, **kwargs
+        )
+
+    @classmethod
+    def add_custom_model(
+        cls,
+        model: str,
+        sources: ModelSource,
+        model_file: str = "onnx/model.onnx",
+        description: str = "",
+        license: str = "",
+        size_in_gb: float = 0.0,
+        additional_files: Optional[list[str]] = None,
+    ) -> None:
+        registered_models = cls._list_supported_models()
+        for registered_model in registered_models:
+            if model == registered_model.model:
+                raise ValueError(
+                    f"Model {model} is already registered in CrossEncoderModel, if you still want to add this model, "
+                    f"please use another model name"
+                )
+
+        CustomTextCrossEncoder.add_model(
+            BaseModelDescription(
+                model=model,
+                sources=sources,
+                model_file=model_file,
+                description=description,
+                license=license,
+                size_in_GB=size_in_gb,
+                additional_files=additional_files or [],
+            )
+        )
@@ -0,0 +1,59 @@
+from typing import Any, Iterable, Optional
+
+from fastembed.common.model_description import BaseModelDescription
+from fastembed.common.model_management import ModelManagement
+
+
+class TextCrossEncoderBase(ModelManagement[BaseModelDescription]):
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        **kwargs: Any,
+    ):
+        self.model_name = model_name
+        self.cache_dir = cache_dir
+        self.threads = threads
+        self._local_files_only = kwargs.pop("local_files_only", False)
+
+    def rerank(
+        self,
+        query: str,
+        documents: Iterable[str],
+        batch_size: int = 64,
+        **kwargs: Any,
+    ) -> Iterable[float]:
+        """Rerank a list of documents given a query.
+
+        Args:
+            query (str): The query to rerank the documents.
+            documents (Iterable[str]): The list of texts to rerank.
+            batch_size (int): The batch size to use for reranking.
+            **kwargs: Additional keyword argument to pass to the rerank method.
+
+        Yields:
+            Iterable[float]: The scores of the reranked the documents.
+        """
+        raise NotImplementedError("This method should be overridden by subclasses")
+
+    def rerank_pairs(
+        self,
+        pairs: Iterable[tuple[str, str]],
+        batch_size: int = 64,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[float]:
+        """Rerank query-document pairs.
+        Args:
+            pairs (Iterable[tuple[str, str]]): Query-document pairs to rerank
+            batch_size (int): The batch size to use for reranking.
+            parallel: parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+            **kwargs: Additional keyword argument to pass to the rerank method.
+        Yields:
+            Iterable[float]: Scores for each individual pair
+        """
+        raise NotImplementedError("This method should be overridden by subclasses")
@@ -0,0 +1,4 @@
+from fastembed.sparse.sparse_embedding_base import SparseEmbedding
+from fastembed.sparse.sparse_text_embedding import SparseTextEmbedding
+
+__all__ = ["SparseEmbedding", "SparseTextEmbedding"]
@@ -0,0 +1,355 @@
+import os
+from collections import defaultdict
+from multiprocessing import get_all_start_methods
+from pathlib import Path
+from typing import Any, Iterable, Optional, Type, Union
+
+import mmh3
+import numpy as np
+from py_rust_stemmers import SnowballStemmer
+from fastembed.common.utils import (
+    define_cache_dir,
+    iter_batch,
+    get_all_punctuation,
+    remove_non_alphanumeric,
+)
+from fastembed.parallel_processor import ParallelWorkerPool, Worker
+from fastembed.sparse.sparse_embedding_base import (
+    SparseEmbedding,
+    SparseTextEmbeddingBase,
+)
+from fastembed.sparse.utils.tokenizer import SimpleTokenizer
+from fastembed.common.model_description import SparseModelDescription, ModelSource
+
+supported_languages = [
+    "arabic",
+    "azerbaijani",
+    "basque",
+    "bengali",
+    "catalan",
+    "chinese",
+    "danish",
+    "dutch",
+    "english",
+    "finnish",
+    "french",
+    "german",
+    "greek",
+    "hebrew",
+    "hinglish",
+    "hungarian",
+    "indonesian",
+    "italian",
+    "kazakh",
+    "nepali",
+    "norwegian",
+    "portuguese",
+    "romanian",
+    "russian",
+    "slovene",
+    "spanish",
+    "swedish",
+    "tajik",
+    "turkish",
+]
+
+supported_bm25_models: list[SparseModelDescription] = [
+    SparseModelDescription(
+        model="Qdrant/bm25",
+        vocab_size=0,
+        description="BM25 as sparse embeddings meant to be used with Qdrant",
+        license="apache-2.0",
+        size_in_GB=0.01,
+        sources=ModelSource(hf="Qdrant/bm25"),
+        additional_files=[f"{lang}.txt" for lang in supported_languages],
+        requires_idf=True,
+        model_file="mock.file",
+    ),
+]
+
+
+class Bm25(SparseTextEmbeddingBase):
+    """Implements traditional BM25 in a form of sparse embeddings.
+    Uses a count of tokens in the document to evaluate the importance of the token.
+
+    WARNING: This model is expected to be used with `modifier="idf"` in the sparse vector index of Qdrant.
+
+    BM25 formula:
+
+    score(q, d) = SUM[ IDF(q_i) * (f(q_i, d) * (k + 1)) / (f(q_i, d) + k * (1 - b + b * (|d| / avg_len))) ],
+
+    where IDF is the inverse document frequency, computed on Qdrant's side
+    f(q_i, d) is the term frequency of the token q_i in the document d
+    k, b, avg_len are hyperparameters, described below.
+
+    Args:
+        model_name (str): The name of the model to use.
+        cache_dir (str, optional): The path to the cache directory.
+            Can be set using the `FASTEMBED_CACHE_PATH` env variable.
+            Defaults to `fastembed_cache` in the system's temp directory.
+        k (float, optional): The k parameter in the BM25 formula. Defines the saturation of the term frequency.
+            I.e. defines how fast the moment when additional terms stop to increase the score. Defaults to 1.2.
+        b (float, optional): The b parameter in the BM25 formula. Defines the importance of the document length.
+            Defaults to 0.75.
+        avg_len (float, optional): The average length of the documents in the corpus. Defaults to 256.0.
+        language (str): Specifies the language for the stemmer.
+        disable_stemmer (bool): Disable the stemmer.
+    Raises:
+        ValueError: If the model_name is not in the format <org>/<model> e.g. BAAI/bge-base-en.
+    """
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        k: float = 1.2,
+        b: float = 0.75,
+        avg_len: float = 256.0,
+        language: str = "english",
+        token_max_length: int = 40,
+        disable_stemmer: bool = False,
+        specific_model_path: Optional[str] = None,
+        **kwargs: Any,
+    ):
+        super().__init__(model_name, cache_dir, **kwargs)
+
+        if language not in supported_languages:
+            raise ValueError(f"{language} language is not supported")
+        else:
+            self.language = language
+
+        self.k = k
+        self.b = b
+        self.avg_len = avg_len
+
+        model_description = self._get_model_description(model_name)
+        self.cache_dir = str(define_cache_dir(cache_dir))
+
+        self._model_dir = self.download_model(
+            model_description,
+            self.cache_dir,
+            local_files_only=self._local_files_only,
+            specific_model_path=specific_model_path,
+        )
+
+        self.token_max_length = token_max_length
+        self.punctuation = set(get_all_punctuation())
+        self.disable_stemmer = disable_stemmer
+
+        if disable_stemmer:
+            self.stopwords: set[str] = set()
+            self.stemmer = None
+        else:
+            self.stopwords = set(self._load_stopwords(self._model_dir, self.language))
+            self.stemmer = SnowballStemmer(language)
+
+        self.tokenizer = SimpleTokenizer
+
+    @classmethod
+    def _list_supported_models(cls) -> list[SparseModelDescription]:
+        """Lists the supported models.
+
+        Returns:
+            list[SparseModelDescription]: A list of SparseModelDescription objects containing the model information.
+        """
+        return supported_bm25_models
+
+    @classmethod
+    def _load_stopwords(cls, model_dir: Path, language: str) -> list[str]:
+        stopwords_path = model_dir / f"{language}.txt"
+        if not stopwords_path.exists():
+            return []
+
+        with open(stopwords_path, "r") as f:
+            return f.read().splitlines()
+
+    def _embed_documents(
+        self,
+        model_name: str,
+        cache_dir: str,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+    ) -> Iterable[SparseEmbedding]:
+        is_small = False
+
+        if isinstance(documents, str):
+            documents = [documents]
+            is_small = True
+
+        if isinstance(documents, list):
+            if len(documents) < batch_size:
+                is_small = True
+
+        if parallel is None or is_small:
+            for batch in iter_batch(documents, batch_size):
+                yield from self.raw_embed(batch)
+        else:
+            if parallel == 0:
+                parallel = os.cpu_count()
+
+            start_method = "forkserver" if "forkserver" in get_all_start_methods() else "spawn"
+            params = {
+                "model_name": model_name,
+                "cache_dir": cache_dir,
+                "k": self.k,
+                "b": self.b,
+                "avg_len": self.avg_len,
+                "language": self.language,
+                "token_max_length": self.token_max_length,
+                "disable_stemmer": self.disable_stemmer,
+            }
+            pool = ParallelWorkerPool(
+                num_workers=parallel or 1,
+                worker=self._get_worker_class(),
+                start_method=start_method,
+            )
+            for batch in pool.ordered_map(iter_batch(documents, batch_size), **params):
+                for record in batch:
+                    yield record  # type: ignore
+
+    def embed(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[SparseEmbedding]:
+        """
+        Encode a list of documents into list of embeddings.
+        We use mean pooling with attention so that the model can handle variable-length inputs.
+
+        Args:
+            documents: Iterator of documents or single document to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per document
+        """
+        yield from self._embed_documents(
+            model_name=self.model_name,
+            cache_dir=str(self.cache_dir),
+            documents=documents,
+            batch_size=batch_size,
+            parallel=parallel,
+        )
+
+    def _stem(self, tokens: list[str]) -> list[str]:
+        stemmed_tokens: list[str] = []
+        for token in tokens:
+            lower_token = token.lower()
+
+            if token in self.punctuation:
+                continue
+
+            if lower_token in self.stopwords:
+                continue
+
+            if len(token) > self.token_max_length:
+                continue
+
+            stemmed_token = self.stemmer.stem_word(lower_token) if self.stemmer else lower_token
+
+            if stemmed_token:
+                stemmed_tokens.append(stemmed_token)
+        return stemmed_tokens
+
+    def raw_embed(
+        self,
+        documents: list[str],
+    ) -> list[SparseEmbedding]:
+        embeddings: list[SparseEmbedding] = []
+        for document in documents:
+            document = remove_non_alphanumeric(document)
+            tokens = self.tokenizer.tokenize(document)
+            stemmed_tokens = self._stem(tokens)
+            token_id2value = self._term_frequency(stemmed_tokens)
+            embeddings.append(SparseEmbedding.from_dict(token_id2value))
+        return embeddings
+
+    def _term_frequency(self, tokens: list[str]) -> dict[int, float]:
+        """Calculate the term frequency part of the BM25 formula.
+
+        (
+            f(q_i, d) * (k + 1)
+        ) / (
+            f(q_i, d) + k * (1 - b + b * (|d| / avg_len))
+        )
+
+        Args:
+            tokens (list[str]): The list of tokens in the document.
+
+        Returns:
+            dict[int, float]: The token_id to term frequency mapping.
+        """
+        tf_map: dict[int, float] = {}
+        counter: defaultdict[str, int] = defaultdict(int)
+        for stemmed_token in tokens:
+            counter[stemmed_token] += 1
+
+        doc_len = len(tokens)
+        for stemmed_token in counter:
+            token_id = self.compute_token_id(stemmed_token)
+            num_occurrences = counter[stemmed_token]
+            tf_map[token_id] = num_occurrences * (self.k + 1)
+            tf_map[token_id] /= num_occurrences + self.k * (
+                1 - self.b + self.b * doc_len / self.avg_len
+            )
+        return tf_map
+
+    @classmethod
+    def compute_token_id(cls, token: str) -> int:
+        return abs(mmh3.hash(token))
+
+    def query_embed(
+        self, query: Union[str, Iterable[str]], **kwargs: Any
+    ) -> Iterable[SparseEmbedding]:
+        """To emulate BM25 behaviour, we don't need to use weights in the query, and
+        it's enough to just hash the tokens and assign a weight of 1.0 to them.
+        """
+        if isinstance(query, str):
+            query = [query]
+
+        for text in query:
+            text = remove_non_alphanumeric(text)
+            tokens = self.tokenizer.tokenize(text)
+            stemmed_tokens = self._stem(tokens)
+            token_ids = np.array(
+                list(set(self.compute_token_id(token) for token in stemmed_tokens)),
+                dtype=np.int32,
+            )
+            values = np.ones_like(token_ids)
+            yield SparseEmbedding(indices=token_ids, values=values)
+
+    @classmethod
+    def _get_worker_class(cls) -> Type["Bm25Worker"]:
+        return Bm25Worker
+
+
+class Bm25Worker(Worker):
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ):
+        self.model = self.init_embedding(model_name, cache_dir, **kwargs)
+
+    @classmethod
+    def start(cls, model_name: str, cache_dir: str, **kwargs: Any) -> "Bm25Worker":
+        return cls(model_name=model_name, cache_dir=cache_dir, **kwargs)
+
+    def process(
+        self, items: Iterable[tuple[int, Any]]
+    ) -> Iterable[tuple[int, list[SparseEmbedding]]]:
+        for idx, batch in items:
+            onnx_output = self.model.raw_embed(batch)
+            yield idx, onnx_output
+
+    @staticmethod
+    def init_embedding(model_name: str, cache_dir: str, **kwargs: Any) -> Bm25:
+        return Bm25(model_name=model_name, cache_dir=cache_dir, **kwargs)
@@ -0,0 +1,346 @@
+import math
+import string
+from pathlib import Path
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+
+import mmh3
+import numpy as np
+from py_rust_stemmers import SnowballStemmer
+
+from fastembed.common import OnnxProvider
+from fastembed.common.onnx_model import OnnxOutputContext
+from fastembed.common.utils import define_cache_dir
+from fastembed.sparse.sparse_embedding_base import (
+    SparseEmbedding,
+    SparseTextEmbeddingBase,
+)
+from fastembed.text.onnx_text_model import OnnxTextModel, TextEmbeddingWorker
+from fastembed.common.model_description import SparseModelDescription, ModelSource
+
+supported_bm42_models: list[SparseModelDescription] = [
+    SparseModelDescription(
+        model="Qdrant/bm42-all-minilm-l6-v2-attentions",
+        vocab_size=30522,
+        description="Light sparse embedding model, which assigns an importance score to each token in the text",
+        license="apache-2.0",
+        size_in_GB=0.09,
+        sources=ModelSource(hf="Qdrant/all_miniLM_L6_v2_with_attentions"),
+        model_file="model.onnx",
+        additional_files=["stopwords.txt"],
+        requires_idf=True,
+    ),
+]
+
+MODEL_TO_LANGUAGE = {
+    "Qdrant/bm42-all-minilm-l6-v2-attentions": "english",
+}
+
+
+class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
+    """
+    Bm42 is an extension of BM25, which tries to better evaluate importance of tokens in the documents,
+    by extracting attention weights from the transformer model.
+
+    Traditional BM25 uses a count of tokens in the document to evaluate the importance of the token,
+    but this approach doesn't work well with short documents or chunks of text, as almost all tokens
+    there are unique.
+
+    BM42 addresses this issue by replacing the token count with the attention weights from the transformer model.
+    This allows sparse embeddings to work well with short documents, handle rare tokens and leverage traditional NLP
+    techniques like stemming and stopwords.
+
+    WARNING: This model is expected to be used with `modifier="idf"` in the sparse vector index of Qdrant.
+    """
+
+    ONNX_OUTPUT_NAMES = ["attention_6"]
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        alpha: float = 0.5,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        device_id: Optional[int] = None,
+        specific_model_path: Optional[str] = None,
+        **kwargs: Any,
+    ):
+        """
+        Args:
+            model_name (str): The name of the model to use.
+            cache_dir (str, optional): The path to the cache directory.
+                                       Can be set using the `FASTEMBED_CACHE_PATH` env variable.
+                                       Defaults to `fastembed_cache` in the system's temp directory.
+            threads (int, optional): The number of threads single onnxruntime session can use. Defaults to None.
+            providers (Optional[Sequence[OnnxProvider]], optional): The providers to use for onnxruntime.
+            alpha (float, optional): Parameter, that defines the importance of the token weight in the document
+                versus the importance of the token frequency in the corpus. Defaults to 0.5, based on empirical testing.
+                It is recommended to only change this parameter based on training data for a specific dataset.
+            cuda (bool, optional): Whether to use cuda for inference. Mutually exclusive with `providers`
+                Defaults to False.
+            device_ids (Optional[list[int]], optional): The list of device ids to use for data parallel processing in
+                workers. Should be used with `cuda=True`, mutually exclusive with `providers`. Defaults to None.
+            lazy_load (bool, optional): Whether to load the model during class initialization or on demand.
+                Should be set to True when using multiple-gpu and parallel encoding. Defaults to False.
+            device_id (Optional[int], optional): The device id to use for loading the model in the worker process.
+            specific_model_path (Optional[str], optional): The specific path to the onnx model dir if it should be imported from somewhere else
+
+        Raises:
+            ValueError: If the model_name is not in the format <org>/<model> e.g. BAAI/bge-base-en.
+        """
+
+        super().__init__(model_name, cache_dir, threads, **kwargs)
+        self.providers = providers
+        self.lazy_load = lazy_load
+
+        # List of device ids, that can be used for data parallel processing in workers
+        self.device_ids = device_ids
+        self.cuda = cuda
+
+        # This device_id will be used if we need to load model in current process
+        self.device_id: Optional[int] = None
+        if device_id is not None:
+            self.device_id = device_id
+        elif self.device_ids is not None:
+            self.device_id = self.device_ids[0]
+
+        self.model_description = self._get_model_description(model_name)
+        self.cache_dir = str(define_cache_dir(cache_dir))
+
+        self._model_dir = self.download_model(
+            self.model_description,
+            self.cache_dir,
+            local_files_only=self._local_files_only,
+            specific_model_path=specific_model_path,
+        )
+
+        self.invert_vocab: dict[int, str] = {}
+
+        self.special_tokens: set[str] = set()
+        self.special_tokens_ids: set[int] = set()
+        self.punctuation = set(string.punctuation)
+        self.stopwords = set(self._load_stopwords(self._model_dir))
+        self.stemmer = SnowballStemmer(MODEL_TO_LANGUAGE[model_name])
+        self.alpha = alpha
+
+        if not self.lazy_load:
+            self.load_onnx_model()
+
+    def load_onnx_model(self) -> None:
+        self._load_onnx_model(
+            model_dir=self._model_dir,
+            model_file=self.model_description.model_file,
+            threads=self.threads,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_id=self.device_id,
+        )
+
+        for token, idx in self.tokenizer.get_vocab().items():  # type: ignore[union-attr]
+            self.invert_vocab[idx] = token
+        self.special_tokens = set(self.special_token_to_id.keys())
+        self.special_tokens_ids = set(self.special_token_to_id.values())
+        self.stopwords = set(self._load_stopwords(self._model_dir))
+
+    def _filter_pair_tokens(self, tokens: list[tuple[str, Any]]) -> list[tuple[str, Any]]:
+        result: list[tuple[str, Any]] = []
+        for token, value in tokens:
+            if token in self.stopwords or token in self.punctuation:
+                continue
+            result.append((token, value))
+        return result
+
+    def _stem_pair_tokens(self, tokens: list[tuple[str, Any]]) -> list[tuple[str, Any]]:
+        result: list[tuple[str, Any]] = []
+        for token, value in tokens:
+            processed_token = self.stemmer.stem_word(token)
+            result.append((processed_token, value))
+        return result
+
+    @classmethod
+    def _aggregate_weights(
+        cls, tokens: list[tuple[str, list[int]]], weights: list[float]
+    ) -> list[tuple[str, float]]:
+        result: list[tuple[str, float]] = []
+        for token, idxs in tokens:
+            sum_weight = sum(weights[idx] for idx in idxs)
+            result.append((token, sum_weight))
+        return result
+
+    def _reconstruct_bpe(
+        self, bpe_tokens: Iterable[tuple[int, str]]
+    ) -> list[tuple[str, list[int]]]:
+        result: list[tuple[str, list[int]]] = []
+        acc: str = ""
+        acc_idx: list[int] = []
+
+        continuing_subword_prefix = self.tokenizer.model.continuing_subword_prefix  # type: ignore[union-attr]
+        continuing_subword_prefix_len = len(continuing_subword_prefix)
+
+        for idx, token in bpe_tokens:
+            if token in self.special_tokens:
+                continue
+
+            if token.startswith(continuing_subword_prefix):
+                acc += token[continuing_subword_prefix_len:]
+                acc_idx.append(idx)
+            else:
+                if acc:
+                    result.append((acc, acc_idx))
+                    acc_idx = []
+                acc = token
+                acc_idx.append(idx)
+
+        if acc:
+            result.append((acc, acc_idx))
+
+        return result
+
+    def _rescore_vector(self, vector: dict[str, float]) -> dict[int, float]:
+        """
+        Orders all tokens in the vector by their importance and generates a new score based on the importance order.
+        So that the scoring doesn't depend on absolute values assigned by the model, but on the relative importance.
+        """
+
+        new_vector: dict[int, float] = {}
+
+        for token, value in vector.items():
+            token_id = abs(mmh3.hash(token))
+            # Examples:
+            # Num 0: Log(1/1 + 1) = 0.6931471805599453
+            # Num 1: Log(1/2 + 1) = 0.4054651081081644
+            # Num 2: Log(1/3 + 1) = 0.28768207245178085
+            new_vector[token_id] = math.log(1.0 + value) ** self.alpha  # value
+
+        return new_vector
+
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[SparseEmbedding]:
+        if output.input_ids is None:
+            raise ValueError("input_ids must be provided for document post-processing")
+
+        token_ids_batch = output.input_ids.astype(int)
+
+        # attention_value shape: (batch_size, num_heads, num_tokens, num_tokens)
+        pooled_attention = np.mean(output.model_output[:, :, 0], axis=1) * output.attention_mask
+
+        for document_token_ids, attention_value in zip(token_ids_batch, pooled_attention):
+            document_tokens_with_ids = (
+                (idx, self.invert_vocab[token_id])
+                for idx, token_id in enumerate(document_token_ids)
+            )
+
+            reconstructed = self._reconstruct_bpe(document_tokens_with_ids)
+
+            filtered = self._filter_pair_tokens(reconstructed)
+
+            stemmed = self._stem_pair_tokens(filtered)
+
+            weighted = self._aggregate_weights(stemmed, attention_value)
+
+            max_token_weight: dict[str, float] = {}
+
+            for token, weight in weighted:
+                max_token_weight[token] = max(max_token_weight.get(token, 0), weight)
+
+            rescored = self._rescore_vector(max_token_weight)
+
+            yield SparseEmbedding.from_dict(rescored)
+
+    @classmethod
+    def _list_supported_models(cls) -> list[SparseModelDescription]:
+        """Lists the supported models.
+
+        Returns:
+            list[SparseModelDescription]: A list of SparseModelDescription objects containing the model information.
+        """
+        return supported_bm42_models
+
+    @classmethod
+    def _load_stopwords(cls, model_dir: Path) -> list[str]:
+        stopwords_path = model_dir / "stopwords.txt"
+        if not stopwords_path.exists():
+            return []
+
+        with open(stopwords_path, "r") as f:
+            return f.read().splitlines()
+
+    def embed(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[SparseEmbedding]:
+        """
+        Encode a list of documents into list of embeddings.
+        We use mean pooling with attention so that the model can handle variable-length inputs.
+
+        Args:
+            documents: Iterator of documents or single document to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per document
+        """
+        yield from self._embed_documents(
+            model_name=self.model_name,
+            cache_dir=str(self.cache_dir),
+            documents=documents,
+            batch_size=batch_size,
+            parallel=parallel,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_ids=self.device_ids,
+            alpha=self.alpha,
+        )
+
+    @classmethod
+    def _query_rehash(cls, tokens: Iterable[str]) -> dict[int, float]:
+        result: dict[int, float] = {}
+        for token in tokens:
+            token_id = abs(mmh3.hash(token))
+            result[token_id] = 1.0
+        return result
+
+    def query_embed(
+        self, query: Union[str, Iterable[str]], **kwargs: Any
+    ) -> Iterable[SparseEmbedding]:
+        """
+        To emulate BM25 behaviour, we don't need to use smart weights in the query, and
+        it's enough to just hash the tokens and assign a weight of 1.0 to them.
+        It is also faster, as we don't need to run the model for the query.
+        """
+        if isinstance(query, str):
+            query = [query]
+
+        if not hasattr(self, "model") or self.model is None:
+            self.load_onnx_model()
+
+        for text in query:
+            encoded = self.tokenizer.encode(text)  # type: ignore[union-attr]
+            document_tokens_with_ids = enumerate(encoded.tokens)
+            reconstructed = self._reconstruct_bpe(document_tokens_with_ids)
+            filtered = self._filter_pair_tokens(reconstructed)
+            stemmed = self._stem_pair_tokens(filtered)
+
+            yield SparseEmbedding.from_dict(self._query_rehash(token for token, _ in stemmed))
+
+    @classmethod
+    def _get_worker_class(cls) -> Type[TextEmbeddingWorker[SparseEmbedding]]:
+        return Bm42TextEmbeddingWorker
+
+
+class Bm42TextEmbeddingWorker(TextEmbeddingWorker[SparseEmbedding]):
+    def init_embedding(self, model_name: str, cache_dir: str, **kwargs: Any) -> Bm42:
+        return Bm42(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            **kwargs,
+        )
@@ -0,0 +1,88 @@
+from dataclasses import dataclass
+from typing import Iterable, Optional, Union, Any
+
+import numpy as np
+from numpy.typing import NDArray
+
+from fastembed.common.model_description import SparseModelDescription
+from fastembed.common.types import NumpyArray
+from fastembed.common.model_management import ModelManagement
+
+
+@dataclass
+class SparseEmbedding:
+    values: NumpyArray
+    indices: Union[NDArray[np.int64], NDArray[np.int32]]
+
+    def as_object(self) -> dict[str, NumpyArray]:
+        return {
+            "values": self.values,
+            "indices": self.indices,
+        }
+
+    def as_dict(self) -> dict[int, float]:
+        return {int(i): float(v) for i, v in zip(self.indices, self.values)}  # type: ignore
+
+    @classmethod
+    def from_dict(cls, data: dict[int, float]) -> "SparseEmbedding":
+        if len(data) == 0:
+            return cls(values=np.array([]), indices=np.array([]))
+        indices, values = zip(*data.items())
+        return cls(values=np.array(values), indices=np.array(indices))
+
+
+class SparseTextEmbeddingBase(ModelManagement[SparseModelDescription]):
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        **kwargs: Any,
+    ):
+        self.model_name = model_name
+        self.cache_dir = cache_dir
+        self.threads = threads
+        self._local_files_only = kwargs.pop("local_files_only", False)
+
+    def embed(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[SparseEmbedding]:
+        raise NotImplementedError()
+
+    def passage_embed(self, texts: Iterable[str], **kwargs: Any) -> Iterable[SparseEmbedding]:
+        """
+        Embeds a list of text passages into a list of embeddings.
+
+        Args:
+            texts (Iterable[str]): The list of texts to embed.
+            **kwargs: Additional keyword argument to pass to the embed method.
+
+        Yields:
+            Iterable[SparseEmbedding]: The sparse embeddings.
+        """
+
+        # This is model-specific, so that different models can have specialized implementations
+        yield from self.embed(texts, **kwargs)
+
+    def query_embed(
+        self, query: Union[str, Iterable[str]], **kwargs: Any
+    ) -> Iterable[SparseEmbedding]:
+        """
+        Embeds queries
+
+        Args:
+            query (Union[str, Iterable[str]]): The query to embed, or an iterable e.g. list of queries.
+
+        Returns:
+            Iterable[SparseEmbedding]: The sparse embeddings.
+        """
+
+        # This is model-specific, so that different models can have specialized implementations
+        if isinstance(query, str):
+            yield from self.embed([query], **kwargs)
+        else:
+            yield from self.embed(query, **kwargs)
@@ -0,0 +1,129 @@
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+from dataclasses import asdict
+
+from fastembed.common import OnnxProvider
+from fastembed.sparse.bm25 import Bm25
+from fastembed.sparse.bm42 import Bm42
+from fastembed.sparse.sparse_embedding_base import (
+    SparseEmbedding,
+    SparseTextEmbeddingBase,
+)
+from fastembed.sparse.splade_pp import SpladePP
+import warnings
+from fastembed.common.model_description import SparseModelDescription
+
+
+class SparseTextEmbedding(SparseTextEmbeddingBase):
+    EMBEDDINGS_REGISTRY: list[Type[SparseTextEmbeddingBase]] = [SpladePP, Bm42, Bm25]
+
+    @classmethod
+    def list_supported_models(cls) -> list[dict[str, Any]]:
+        """
+        Lists the supported models.
+
+        Returns:
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
+
+            Example:
+                ```
+                [
+                    {
+                        "model": "prithvida/SPLADE_PP_en_v1",
+                        "vocab_size": 30522,
+                        "description": "Independent Implementation of SPLADE++ Model for English",
+                        "license": "apache-2.0",
+                        "size_in_GB": 0.532,
+                        "sources": {
+                            "hf": "qdrant/SPLADE_PP_en_v1",
+                        },
+                    }
+                ]
+                ```
+        """
+        return [asdict(model) for model in cls._list_supported_models()]
+
+    @classmethod
+    def _list_supported_models(cls) -> list[SparseModelDescription]:
+        result: list[SparseModelDescription] = []
+        for embedding in cls.EMBEDDINGS_REGISTRY:
+            result.extend(embedding._list_supported_models())
+        return result
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        **kwargs: Any,
+    ):
+        super().__init__(model_name, cache_dir, threads, **kwargs)
+        if model_name == "prithvida/Splade_PP_en_v1":
+            warnings.warn(
+                "The right spelling is prithivida/Splade_PP_en_v1. "
+                "Support of this name will be removed soon, please fix the model_name",
+                DeprecationWarning,
+                stacklevel=2,
+            )
+            model_name = "prithivida/Splade_PP_en_v1"
+
+        for EMBEDDING_MODEL_TYPE in self.EMBEDDINGS_REGISTRY:
+            supported_models = EMBEDDING_MODEL_TYPE._list_supported_models()
+            if any(model_name.lower() == model.model.lower() for model in supported_models):
+                self.model = EMBEDDING_MODEL_TYPE(
+                    model_name,
+                    cache_dir,
+                    threads=threads,
+                    providers=providers,
+                    cuda=cuda,
+                    device_ids=device_ids,
+                    lazy_load=lazy_load,
+                    **kwargs,
+                )
+                return
+
+        raise ValueError(
+            f"Model {model_name} is not supported in SparseTextEmbedding."
+            "Please check the supported models using `SparseTextEmbedding.list_supported_models()`"
+        )
+
+    def embed(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[SparseEmbedding]:
+        """
+        Encode a list of documents into list of embeddings.
+        We use mean pooling with attention so that the model can handle variable-length inputs.
+
+        Args:
+            documents: Iterator of documents or single document to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per document
+        """
+        yield from self.model.embed(documents, batch_size, parallel, **kwargs)
+
+    def query_embed(
+        self, query: Union[str, Iterable[str]], **kwargs: Any
+    ) -> Iterable[SparseEmbedding]:
+        """
+        Embeds queries
+
+        Args:
+            query (Union[str, Iterable[str]]): The query to embed, or an iterable e.g. list of queries.
+
+        Returns:
+            Iterable[SparseEmbedding]: The sparse embeddings.
+        """
+        yield from self.model.query_embed(query, **kwargs)
@@ -0,0 +1,181 @@
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+
+import numpy as np
+from fastembed.common import OnnxProvider
+from fastembed.common.onnx_model import OnnxOutputContext
+from fastembed.common.utils import define_cache_dir
+from fastembed.sparse.sparse_embedding_base import (
+    SparseEmbedding,
+    SparseTextEmbeddingBase,
+)
+from fastembed.text.onnx_text_model import OnnxTextModel, TextEmbeddingWorker
+from fastembed.common.model_description import SparseModelDescription, ModelSource
+
+supported_splade_models: list[SparseModelDescription] = [
+    SparseModelDescription(
+        model="prithivida/Splade_PP_en_v1",
+        vocab_size=30522,
+        description="Independent Implementation of SPLADE++ Model for English.",
+        license="apache-2.0",
+        size_in_GB=0.532,
+        sources=ModelSource(hf="Qdrant/Splade_PP_en_v1"),
+        model_file="model.onnx",
+    ),
+    SparseModelDescription(
+        model="prithvida/Splade_PP_en_v1",
+        vocab_size=30522,
+        description="Independent Implementation of SPLADE++ Model for English.",
+        license="apache-2.0",
+        size_in_GB=0.532,
+        sources=ModelSource(hf="Qdrant/Splade_PP_en_v1"),
+        model_file="model.onnx",
+    ),
+]
+
+
+class SpladePP(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[SparseEmbedding]:
+        if output.attention_mask is None:
+            raise ValueError("attention_mask must be provided for document post-processing")
+
+        relu_log = np.log(1 + np.maximum(output.model_output, 0))
+
+        weighted_log = relu_log * np.expand_dims(output.attention_mask, axis=-1)
+
+        scores = np.max(weighted_log, axis=1)
+
+        # Score matrix of shape (batch_size, vocab_size)
+        # Most of the values are 0, only a few are non-zero
+        for row_scores in scores:
+            indices = row_scores.nonzero()[0]
+            scores = row_scores[indices]
+            yield SparseEmbedding(values=scores, indices=indices)
+
+    @classmethod
+    def _list_supported_models(cls) -> list[SparseModelDescription]:
+        """Lists the supported models.
+
+        Returns:
+            list[SparseModelDescription]: A list of SparseModelDescription objects containing the model information.
+        """
+        return supported_splade_models
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        device_id: Optional[int] = None,
+        specific_model_path: Optional[str] = None,
+        **kwargs: Any,
+    ):
+        """
+        Args:
+            model_name (str): The name of the model to use.
+            cache_dir (str, optional): The path to the cache directory.
+                                       Can be set using the `FASTEMBED_CACHE_PATH` env variable.
+                                       Defaults to `fastembed_cache` in the system's temp directory.
+            threads (int, optional): The number of threads single onnxruntime session can use. Defaults to None.
+            providers (Optional[Sequence[OnnxProvider]], optional): The list of onnxruntime providers to use.
+                Mutually exclusive with the `cuda` and `device_ids` arguments. Defaults to None.
+            cuda (bool, optional): Whether to use cuda for inference. Mutually exclusive with `providers`
+                Defaults to False.
+            device_ids (Optional[list[int]], optional): The list of device ids to use for data parallel processing in
+                workers. Should be used with `cuda=True`, mutually exclusive with `providers`. Defaults to None.
+            lazy_load (bool, optional): Whether to load the model during class initialization or on demand.
+                Should be set to True when using multiple-gpu and parallel encoding. Defaults to False.
+            device_id (Optional[int], optional): The device id to use for loading the model in the worker process.
+            specific_model_path (Optional[str], optional): The specific path to the onnx model dir if it should be imported from somewhere else
+
+        Raises:
+            ValueError: If the model_name is not in the format <org>/<model> e.g. BAAI/bge-base-en.
+        """
+        super().__init__(model_name, cache_dir, threads, **kwargs)
+        self.providers = providers
+        self.lazy_load = lazy_load
+
+        # List of device ids, that can be used for data parallel processing in workers
+        self.device_ids = device_ids
+        self.cuda = cuda
+
+        # This device_id will be used if we need to load model in current process
+        self.device_id: Optional[int] = None
+        if device_id is not None:
+            self.device_id = device_id
+        elif self.device_ids is not None:
+            self.device_id = self.device_ids[0]
+
+        self.model_description = self._get_model_description(model_name)
+        self.cache_dir = str(define_cache_dir(cache_dir))
+
+        self._model_dir = self.download_model(
+            self.model_description,
+            self.cache_dir,
+            local_files_only=self._local_files_only,
+            specific_model_path=specific_model_path,
+        )
+
+        if not self.lazy_load:
+            self.load_onnx_model()
+
+    def load_onnx_model(self) -> None:
+        self._load_onnx_model(
+            model_dir=self._model_dir,
+            model_file=self.model_description.model_file,
+            threads=self.threads,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_id=self.device_id,
+        )
+
+    def embed(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[SparseEmbedding]:
+        """
+        Encode a list of documents into list of embeddings.
+        We use mean pooling with attention so that the model can handle variable-length inputs.
+
+        Args:
+            documents: Iterator of documents or single document to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per document
+        """
+        yield from self._embed_documents(
+            model_name=self.model_name,
+            cache_dir=str(self.cache_dir),
+            documents=documents,
+            batch_size=batch_size,
+            parallel=parallel,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_ids=self.device_ids,
+            **kwargs,
+        )
+
+    @classmethod
+    def _get_worker_class(cls) -> Type[TextEmbeddingWorker[SparseEmbedding]]:
+        return SpladePPEmbeddingWorker
+
+
+class SpladePPEmbeddingWorker(TextEmbeddingWorker[SparseEmbedding]):
+    def init_embedding(self, model_name: str, cache_dir: str, **kwargs: Any) -> SpladePP:
+        return SpladePP(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )
@@ -0,0 +1,120 @@
+# This code is a modified copy of the `NLTKWordTokenizer` class from `NLTK` library.
+
+import re
+
+
+class SimpleTokenizer:
+    @staticmethod
+    def tokenize(text: str) -> list[str]:
+        text = re.sub(r"[^\w]", " ", text.lower())
+        text = re.sub(r"\s+", " ", text)
+
+        return text.strip().split()
+
+
+class WordTokenizer:
+    """The tokenizer is "destructive" such that the regexes applied will munge the
+    input string to a state beyond re-construction.
+    """
+
+    # Starting quotes.
+    STARTING_QUOTES = [
+        (re.compile("([«“‘„]|[`]+)", re.U), r" \1 "),
+        (re.compile(r"^\""), r"``"),
+        (re.compile(r"(``)"), r" \1 "),
+        (re.compile(r"([ \(\[{<])(\"|\'{2})"), r"\1 `` "),
+        (re.compile(r"(?i)(\')(?!re|ve|ll|m|t|s|d|n)(\w)\b", re.U), r"\1 \2"),
+    ]
+
+    # Ending quotes.
+    ENDING_QUOTES = [
+        (re.compile("([»”’])", re.U), r" \1 "),
+        (re.compile(r"''"), " '' "),
+        (re.compile(r'"'), " '' "),
+        (re.compile(r"([^' ])('[sS]|'[mM]|'[dD]|') "), r"\1 \2 "),
+        (re.compile(r"([^' ])('ll|'LL|'re|'RE|'ve|'VE|n't|N'T) "), r"\1 \2 "),
+    ]
+
+    # Punctuation.
+    PUNCTUATION = [
+        (re.compile(r'([^\.])(\.)([\]\)}>"\'' "»”’ " r"]*)\s*$", re.U), r"\1 \2 \3 "),
+        (re.compile(r"([:,])([^\d])"), r" \1 \2"),
+        (re.compile(r"([:,])$"), r" \1 "),
+        (
+            re.compile(r"\.{2,}", re.U),
+            r" \g<0> ",
+        ),
+        (re.compile(r"[;@#$%&]"), r" \g<0> "),
+        (
+            re.compile(r'([^\.])(\.)([\]\)}>"\']*)\s*$'),
+            r"\1 \2\3 ",
+        ),  # Handles the final period.
+        (re.compile(r"[?!]"), r" \g<0> "),
+        (re.compile(r"([^'])' "), r"\1 ' "),
+        (
+            re.compile(r"[*]", re.U),
+            r" \g<0> ",
+        ),
+    ]
+
+    # Pads parentheses
+    PARENS_BRACKETS = (re.compile(r"[\]\[\(\)\{\}\<\>]"), r" \g<0> ")
+    DOUBLE_DASHES = (re.compile(r"--"), r" -- ")
+
+    # List of contractions adapted from Robert MacIntyre's tokenizer.
+    CONTRACTIONS2 = [
+        re.compile(pattern)
+        for pattern in (
+            r"(?i)\b(can)(?#X)(not)\b",
+            r"(?i)\b(d)(?#X)('ye)\b",
+            r"(?i)\b(gim)(?#X)(me)\b",
+            r"(?i)\b(gon)(?#X)(na)\b",
+            r"(?i)\b(got)(?#X)(ta)\b",
+            r"(?i)\b(lem)(?#X)(me)\b",
+            r"(?i)\b(more)(?#X)('n)\b",
+            r"(?i)\b(wan)(?#X)(na)(?=\s)",
+        )
+    ]
+    CONTRACTIONS3 = [
+        re.compile(pattern) for pattern in (r"(?i) ('t)(?#X)(is)\b", r"(?i) ('t)(?#X)(was)\b")
+    ]
+
+    @classmethod
+    def tokenize(cls, text: str) -> list[str]:
+        """Return a tokenized copy of `text`.
+
+        >>> s = '''Good muffins cost $3.88 (roughly 3,36 euros)\nin New York.'''
+        >>> WordTokenizer().tokenize(s)
+        ['Good', 'muffins', 'cost', '$', '3.88', '(', 'roughly', '3,36', 'euros', ')', 'in', 'New', 'York', '.']
+
+        Args:
+            text: The text to be tokenized.
+
+        Returns:
+            A list of tokens.
+        """
+        for regexp, substitution in cls.STARTING_QUOTES:
+            text = regexp.sub(substitution, text)
+
+        for regexp, substitution in cls.PUNCTUATION:
+            text = regexp.sub(substitution, text)
+
+        # Handles parentheses.
+        regexp, substitution = cls.PARENS_BRACKETS
+        text = regexp.sub(substitution, text)
+
+        # Handles double dash.
+        regexp, substitution = cls.DOUBLE_DASHES
+        text = regexp.sub(substitution, text)
+
+        # add extra space to make things easier
+        text = " " + text + " "
+
+        for regexp, substitution in cls.ENDING_QUOTES:
+            text = regexp.sub(substitution, text)
+
+        for regexp in cls.CONTRACTIONS2:
+            text = regexp.sub(r" \1 \2 ", text)
+        for regexp in cls.CONTRACTIONS3:
+            text = regexp.sub(r" \1 \2 ", text)
+        return text.split()
@@ -0,0 +1,3 @@
+from fastembed.text.text_embedding import TextEmbedding
+
+__all__ = ["TextEmbedding"]
@@ -0,0 +1,54 @@
+from typing import Any, Iterable, Type
+
+from fastembed.common.types import NumpyArray
+from fastembed.common.onnx_model import OnnxOutputContext
+from fastembed.text.onnx_embedding import OnnxTextEmbedding, OnnxTextEmbeddingWorker
+from fastembed.common.model_description import DenseModelDescription, ModelSource
+
+supported_clip_models: list[DenseModelDescription] = [
+    DenseModelDescription(
+        model="Qdrant/clip-ViT-B-32-text",
+        dim=512,
+        description=(
+            "Text embeddings, Multimodal (text&image), English, 77 input tokens truncation, "
+            "Prefixes for queries/documents: not necessary, 2021 year"
+        ),
+        license="mit",
+        size_in_GB=0.25,
+        sources=ModelSource(hf="Qdrant/clip-ViT-B-32-text"),
+        model_file="model.onnx",
+    ),
+]
+
+
+class CLIPOnnxEmbedding(OnnxTextEmbedding):
+    @classmethod
+    def _get_worker_class(cls) -> Type[OnnxTextEmbeddingWorker]:
+        return CLIPEmbeddingWorker
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        """Lists the supported models.
+
+        Returns:
+            list[DenseModelDescription]: A list of DenseModelDescription objects containing the model information.
+        """
+        return supported_clip_models
+
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[NumpyArray]:
+        return output.model_output
+
+
+class CLIPEmbeddingWorker(OnnxTextEmbeddingWorker):
+    def init_embedding(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ) -> OnnxTextEmbedding:
+        return CLIPOnnxEmbedding(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )
@@ -0,0 +1,91 @@
+from typing import Optional, Sequence, Any, Iterable
+
+from dataclasses import dataclass
+
+import numpy as np
+from numpy.typing import NDArray
+
+from fastembed.common import OnnxProvider
+from fastembed.common.model_description import (
+    PoolingType,
+    DenseModelDescription,
+)
+from fastembed.common.onnx_model import OnnxOutputContext
+from fastembed.common.types import NumpyArray
+from fastembed.common.utils import normalize, mean_pooling
+from fastembed.text.onnx_embedding import OnnxTextEmbedding
+
+
+@dataclass(frozen=True)
+class PostprocessingConfig:
+    pooling: PoolingType
+    normalization: bool
+
+
+class CustomTextEmbedding(OnnxTextEmbedding):
+    SUPPORTED_MODELS: list[DenseModelDescription] = []
+    POSTPROCESSING_MAPPING: dict[str, PostprocessingConfig] = {}
+
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        device_id: Optional[int] = None,
+        specific_model_path: Optional[str] = None,
+        **kwargs: Any,
+    ):
+        super().__init__(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=threads,
+            providers=providers,
+            cuda=cuda,
+            device_ids=device_ids,
+            lazy_load=lazy_load,
+            device_id=device_id,
+            specific_model_path=specific_model_path,
+            **kwargs,
+        )
+        self._pooling = self.POSTPROCESSING_MAPPING[model_name].pooling
+        self._normalization = self.POSTPROCESSING_MAPPING[model_name].normalization
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        return cls.SUPPORTED_MODELS
+
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[NumpyArray]:
+        return self._normalize(self._pool(output.model_output, output.attention_mask))
+
+    def _pool(
+        self, embeddings: NumpyArray, attention_mask: Optional[NDArray[np.int64]] = None
+    ) -> NumpyArray:
+        if self._pooling == PoolingType.CLS:
+            return embeddings[:, 0]
+
+        if self._pooling == PoolingType.MEAN:
+            if attention_mask is None:
+                raise ValueError("attention_mask must be provided for mean pooling")
+            return mean_pooling(embeddings, attention_mask)
+
+        if self._pooling == PoolingType.DISABLED:
+            return embeddings
+
+    def _normalize(self, embeddings: NumpyArray) -> NumpyArray:
+        return normalize(embeddings) if self._normalization else embeddings
+
+    @classmethod
+    def add_model(
+        cls,
+        model_description: DenseModelDescription,
+        pooling: PoolingType,
+        normalization: bool,
+    ) -> None:
+        cls.SUPPORTED_MODELS.append(model_description)
+        cls.POSTPROCESSING_MAPPING[model_description.model] = PostprocessingConfig(
+            pooling=pooling, normalization=normalization
+        )
@@ -0,0 +1,111 @@
+from enum import Enum
+from typing import Any, Type, Iterable, Union, Optional
+
+import numpy as np
+
+from fastembed.common.onnx_model import OnnxOutputContext
+from fastembed.common.types import NumpyArray
+from fastembed.text.pooled_normalized_embedding import PooledNormalizedEmbedding
+from fastembed.text.onnx_embedding import OnnxTextEmbeddingWorker
+from fastembed.common.model_description import DenseModelDescription, ModelSource
+
+supported_multitask_models: list[DenseModelDescription] = [
+    DenseModelDescription(
+        model="jinaai/jina-embeddings-v3",
+        dim=1024,
+        tasks={
+            "retrieval.query": 0,
+            "retrieval.passage": 1,
+            "separation": 2,
+            "classification": 3,
+            "text-matching": 4,
+        },
+        description=(
+            "Multi-task unimodal (text) embedding model, multi-lingual (~100), "
+            "1024 tokens truncation, and 8192 sequence length. Prefixes for queries/documents: not necessary, 2024 year."
+        ),
+        license="cc-by-nc-4.0",
+        size_in_GB=2.29,
+        sources=ModelSource(hf="jinaai/jina-embeddings-v3"),
+        model_file="onnx/model.onnx",
+        additional_files=["onnx/model.onnx_data"],
+    ),
+]
+
+
+class Task(int, Enum):
+    RETRIEVAL_QUERY = 0
+    RETRIEVAL_PASSAGE = 1
+    SEPARATION = 2
+    CLASSIFICATION = 3
+    TEXT_MATCHING = 4
+
+
+class JinaEmbeddingV3(PooledNormalizedEmbedding):
+    PASSAGE_TASK = Task.RETRIEVAL_PASSAGE
+    QUERY_TASK = Task.RETRIEVAL_QUERY
+
+    def __init__(self, *args: Any, task_id: Optional[int] = None, **kwargs: Any):
+        super().__init__(*args, **kwargs)
+        self.default_task_id: Union[Task, int] = (
+            task_id if task_id is not None else self.PASSAGE_TASK
+        )
+
+    @classmethod
+    def _get_worker_class(cls) -> Type[OnnxTextEmbeddingWorker]:
+        return JinaEmbeddingV3Worker
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        return supported_multitask_models
+
+    def _preprocess_onnx_input(
+        self,
+        onnx_input: dict[str, NumpyArray],
+        task_id: Optional[Union[int, Task]] = None,
+        **kwargs: Any,
+    ) -> dict[str, NumpyArray]:
+        if task_id is None:
+            raise ValueError(f"task_id must be provided for JinaEmbeddingV3, got <{task_id}>")
+        onnx_input["task_id"] = np.array(task_id, dtype=np.int64)
+        return onnx_input
+
+    def embed(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        task_id: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        task_id = (
+            task_id if task_id is not None else self.default_task_id
+        )  # required for multiprocessing
+        yield from super().embed(documents, batch_size, parallel, task_id=task_id, **kwargs)
+
+    def query_embed(self, query: Union[str, Iterable[str]], **kwargs: Any) -> Iterable[NumpyArray]:
+        yield from super().embed(query, task_id=self.QUERY_TASK, **kwargs)
+
+    def passage_embed(self, texts: Iterable[str], **kwargs: Any) -> Iterable[NumpyArray]:
+        yield from super().embed(texts, task_id=self.PASSAGE_TASK, **kwargs)
+
+
+class JinaEmbeddingV3Worker(OnnxTextEmbeddingWorker):
+    def init_embedding(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ) -> JinaEmbeddingV3:
+        return JinaEmbeddingV3(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )
+
+    def process(self, items: Iterable[tuple[int, Any]]) -> Iterable[tuple[int, OnnxOutputContext]]:
+        self.model: JinaEmbeddingV3  # mypy complaints `self.model` does not have `default_task_id`
+        for idx, batch in items:
+            onnx_output = self.model.onnx_embed(batch, task_id=self.model.default_task_id)
+            yield idx, onnx_output
@@ -0,0 +1,340 @@
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+
+from fastembed.common.types import NumpyArray, OnnxProvider
+from fastembed.common.onnx_model import OnnxOutputContext
+from fastembed.common.utils import define_cache_dir, normalize
+from fastembed.text.onnx_text_model import OnnxTextModel, TextEmbeddingWorker
+from fastembed.text.text_embedding_base import TextEmbeddingBase
+from fastembed.common.model_description import DenseModelDescription, ModelSource
+
+supported_onnx_models: list[DenseModelDescription] = [
+    DenseModelDescription(
+        model="BAAI/bge-base-en",
+        dim=768,
+        description=(
+            "Text embeddings, Unimodal (text), English, 512 input tokens truncation, "
+            "Prefixes for queries/documents: necessary, 2023 year."
+        ),
+        license="mit",
+        size_in_GB=0.42,
+        sources=ModelSource(
+            hf="Qdrant/fast-bge-base-en",
+            url="https://storage.googleapis.com/qdrant-fastembed/fast-bge-base-en.tar.gz",
+            _deprecated_tar_struct=True,
+        ),
+        model_file="model_optimized.onnx",
+    ),
+    DenseModelDescription(
+        model="BAAI/bge-base-en-v1.5",
+        dim=768,
+        description=(
+            "Text embeddings, Unimodal (text), English, 512 input tokens truncation, "
+            "Prefixes for queries/documents: not so necessary, 2023 year."
+        ),
+        license="mit",
+        size_in_GB=0.21,
+        sources=ModelSource(
+            hf="qdrant/bge-base-en-v1.5-onnx-q",
+            url="https://storage.googleapis.com/qdrant-fastembed/fast-bge-base-en-v1.5.tar.gz",
+            _deprecated_tar_struct=True,
+        ),
+        model_file="model_optimized.onnx",
+    ),
+    DenseModelDescription(
+        model="BAAI/bge-large-en-v1.5",
+        dim=1024,
+        description=(
+            "Text embeddings, Unimodal (text), English, 512 input tokens truncation, "
+            "Prefixes for queries/documents: not so necessary, 2023 year."
+        ),
+        license="mit",
+        size_in_GB=1.20,
+        sources=ModelSource(hf="qdrant/bge-large-en-v1.5-onnx"),
+        model_file="model.onnx",
+    ),
+    DenseModelDescription(
+        model="BAAI/bge-small-en",
+        dim=384,
+        description=(
+            "Text embeddings, Unimodal (text), English, 512 input tokens truncation, "
+            "Prefixes for queries/documents: necessary, 2023 year."
+        ),
+        license="mit",
+        size_in_GB=0.13,
+        sources=ModelSource(
+            hf="Qdrant/bge-small-en",
+            url="https://storage.googleapis.com/qdrant-fastembed/BAAI-bge-small-en.tar.gz",
+            _deprecated_tar_struct=True,
+        ),
+        model_file="model_optimized.onnx",
+    ),
+    DenseModelDescription(
+        model="BAAI/bge-small-en-v1.5",
+        dim=384,
+        description=(
+            "Text embeddings, Unimodal (text), English, 512 input tokens truncation, "
+            "Prefixes for queries/documents: not so necessary, 2023 year."
+        ),
+        license="mit",
+        size_in_GB=0.067,
+        sources=ModelSource(hf="qdrant/bge-small-en-v1.5-onnx-q"),
+        model_file="model_optimized.onnx",
+    ),
+    DenseModelDescription(
+        model="BAAI/bge-small-zh-v1.5",
+        dim=512,
+        description=(
+            "Text embeddings, Unimodal (text), Chinese, 512 input tokens truncation, "
+            "Prefixes for queries/documents: not so necessary, 2023 year."
+        ),
+        license="mit",
+        size_in_GB=0.09,
+        sources=ModelSource(
+            hf="Qdrant/bge-small-zh-v1.5",
+            url="https://storage.googleapis.com/qdrant-fastembed/fast-bge-small-zh-v1.5.tar.gz",
+            _deprecated_tar_struct=True,
+        ),
+        model_file="model_optimized.onnx",
+    ),
+    DenseModelDescription(
+        model="mixedbread-ai/mxbai-embed-large-v1",
+        dim=1024,
+        description=(
+            "Text embeddings, Unimodal (text), English, 512 input tokens truncation, "
+            "Prefixes for queries/documents: necessary, 2024 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.64,
+        sources=ModelSource(hf="mixedbread-ai/mxbai-embed-large-v1"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="snowflake/snowflake-arctic-embed-xs",
+        dim=384,
+        description=(
+            "Text embeddings, Unimodal (text), English, 512 input tokens truncation, "
+            "Prefixes for queries/documents: necessary, 2024 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.09,
+        sources=ModelSource(hf="snowflake/snowflake-arctic-embed-xs"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="snowflake/snowflake-arctic-embed-s",
+        dim=384,
+        description=(
+            "Text embeddings, Unimodal (text), English, 512 input tokens truncation, "
+            "Prefixes for queries/documents: necessary, 2024 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.13,
+        sources=ModelSource(hf="snowflake/snowflake-arctic-embed-s"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="snowflake/snowflake-arctic-embed-m",
+        dim=768,
+        description=(
+            "Text embeddings, Unimodal (text), English, 512 input tokens truncation, "
+            "Prefixes for queries/documents: necessary, 2024 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.43,
+        sources=ModelSource(hf="Snowflake/snowflake-arctic-embed-m"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="snowflake/snowflake-arctic-embed-m-long",
+        dim=768,
+        description=(
+            "Text embeddings, Unimodal (text), English, 2048 input tokens truncation, "
+            "Prefixes for queries/documents: necessary, 2024 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.54,
+        sources=ModelSource(hf="snowflake/snowflake-arctic-embed-m-long"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="snowflake/snowflake-arctic-embed-l",
+        dim=1024,
+        description=(
+            "Text embeddings, Unimodal (text), English, 512 input tokens truncation, "
+            "Prefixes for queries/documents: necessary, 2024 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=1.02,
+        sources=ModelSource(hf="snowflake/snowflake-arctic-embed-l"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="jinaai/jina-clip-v1",
+        dim=768,
+        description=(
+            "Text embeddings, Multimodal (text&image), English, Prefixes for queries/documents: "
+            "not necessary, 2024 year"
+        ),
+        license="apache-2.0",
+        size_in_GB=0.55,
+        sources=ModelSource(hf="jinaai/jina-clip-v1"),
+        model_file="onnx/text_model.onnx",
+    ),
+]
+
+
+class OnnxTextEmbedding(TextEmbeddingBase, OnnxTextModel[NumpyArray]):
+    """Implementation of the Flag Embedding model."""
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        """
+        Lists the supported models.
+
+        Returns:
+            list[DenseModelDescription]: A list of DenseModelDescription objects containing the model information.
+        """
+        return supported_onnx_models
+
+    def __init__(
+        self,
+        model_name: str = "BAAI/bge-small-en-v1.5",
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        device_id: Optional[int] = None,
+        specific_model_path: Optional[str] = None,
+        **kwargs: Any,
+    ):
+        """
+        Args:
+            model_name (str): The name of the model to use.
+            cache_dir (str, optional): The path to the cache directory.
+                                       Can be set using the `FASTEMBED_CACHE_PATH` env variable.
+                                       Defaults to `fastembed_cache` in the system's temp directory.
+            threads (int, optional): The number of threads single onnxruntime session can use. Defaults to None.
+            providers (Optional[Sequence[OnnxProvider]], optional): The list of onnxruntime providers to use.
+                Mutually exclusive with the `cuda` and `device_ids` arguments. Defaults to None.
+            cuda (bool, optional): Whether to use cuda for inference. Mutually exclusive with `providers`
+                Defaults to False.
+            device_ids (Optional[list[int]], optional): The list of device ids to use for data parallel processing in
+                workers. Should be used with `cuda=True`, mutually exclusive with `providers`. Defaults to None.
+            lazy_load (bool, optional): Whether to load the model during class initialization or on demand.
+                Should be set to True when using multiple-gpu and parallel encoding. Defaults to False.
+            device_id (Optional[int], optional): The device id to use for loading the model in the worker process.
+            specific_model_path (Optional[str], optional): The specific path to the onnx model dir if it should be imported from somewhere else
+
+        Raises:
+            ValueError: If the model_name is not in the format <org>/<model> e.g. BAAI/bge-base-en.
+        """
+        super().__init__(model_name, cache_dir, threads, **kwargs)
+        self.providers = providers
+        self.lazy_load = lazy_load
+
+        # List of device ids, that can be used for data parallel processing in workers
+        self.device_ids = device_ids
+        self.cuda = cuda
+
+        # This device_id will be used if we need to load model in current process
+        self.device_id: Optional[int] = None
+        if device_id is not None:
+            self.device_id = device_id
+        elif self.device_ids is not None:
+            self.device_id = self.device_ids[0]
+
+        self.model_description = self._get_model_description(model_name)
+        self.cache_dir = str(define_cache_dir(cache_dir))
+        self._model_dir = self.download_model(
+            self.model_description,
+            self.cache_dir,
+            local_files_only=self._local_files_only,
+            specific_model_path=specific_model_path,
+        )
+
+        if not self.lazy_load:
+            self.load_onnx_model()
+
+    def embed(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        """
+        Encode a list of documents into list of embeddings.
+        We use mean pooling with attention so that the model can handle variable-length inputs.
+
+        Args:
+            documents: Iterator of documents or single document to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per document
+        """
+        yield from self._embed_documents(
+            model_name=self.model_name,
+            cache_dir=str(self.cache_dir),
+            documents=documents,
+            batch_size=batch_size,
+            parallel=parallel,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_ids=self.device_ids,
+            **kwargs,
+        )
+
+    @classmethod
+    def _get_worker_class(cls) -> Type["TextEmbeddingWorker[NumpyArray]"]:
+        return OnnxTextEmbeddingWorker
+
+    def _preprocess_onnx_input(
+        self, onnx_input: dict[str, NumpyArray], **kwargs: Any
+    ) -> dict[str, NumpyArray]:
+        """
+        Preprocess the onnx input.
+        """
+        return onnx_input
+
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[NumpyArray]:
+        embeddings = output.model_output
+
+        if embeddings.ndim == 3:  # (batch_size, seq_len, embedding_dim)
+            processed_embeddings = embeddings[:, 0]
+        elif embeddings.ndim == 2:  # (batch_size, embedding_dim)
+            processed_embeddings = embeddings
+        else:
+            raise ValueError(f"Unsupported embedding shape: {embeddings.shape}")
+        return normalize(processed_embeddings)
+
+    def load_onnx_model(self) -> None:
+        self._load_onnx_model(
+            model_dir=self._model_dir,
+            model_file=self.model_description.model_file,
+            threads=self.threads,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_id=self.device_id,
+        )
+
+
+class OnnxTextEmbeddingWorker(TextEmbeddingWorker[NumpyArray]):
+    def init_embedding(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ) -> OnnxTextEmbedding:
+        return OnnxTextEmbedding(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )
@@ -0,0 +1,146 @@
+import os
+from multiprocessing import get_all_start_methods
+from pathlib import Path
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+
+import numpy as np
+from numpy.typing import NDArray
+from tokenizers import Encoding, Tokenizer
+
+from fastembed.common.types import NumpyArray, OnnxProvider
+from fastembed.common.onnx_model import EmbeddingWorker, OnnxModel, OnnxOutputContext, T
+from fastembed.common.preprocessor_utils import load_tokenizer
+from fastembed.common.utils import iter_batch
+from fastembed.parallel_processor import ParallelWorkerPool
+
+
+class OnnxTextModel(OnnxModel[T]):
+    ONNX_OUTPUT_NAMES: Optional[list[str]] = None
+
+    @classmethod
+    def _get_worker_class(cls) -> Type["TextEmbeddingWorker[T]"]:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[T]:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def __init__(self) -> None:
+        super().__init__()
+        self.tokenizer: Optional[Tokenizer] = None
+        self.special_token_to_id: dict[str, int] = {}
+
+    def _preprocess_onnx_input(
+        self, onnx_input: dict[str, NumpyArray], **kwargs: Any
+    ) -> dict[str, Union[NumpyArray, NDArray[np.int64]]]:
+        """
+        Preprocess the onnx input.
+        """
+        return onnx_input
+
+    def _load_onnx_model(
+        self,
+        model_dir: Path,
+        model_file: str,
+        threads: Optional[int],
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_id: Optional[int] = None,
+    ) -> None:
+        super()._load_onnx_model(
+            model_dir=model_dir,
+            model_file=model_file,
+            threads=threads,
+            providers=providers,
+            cuda=cuda,
+            device_id=device_id,
+        )
+        self.tokenizer, self.special_token_to_id = load_tokenizer(model_dir=model_dir)
+
+    def load_onnx_model(self) -> None:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def tokenize(self, documents: list[str], **kwargs: Any) -> list[Encoding]:
+        return self.tokenizer.encode_batch(documents)  # type: ignore[union-attr]
+
+    def onnx_embed(
+        self,
+        documents: list[str],
+        **kwargs: Any,
+    ) -> OnnxOutputContext:
+        encoded = self.tokenize(documents, **kwargs)
+        input_ids = np.array([e.ids for e in encoded])
+        attention_mask = np.array([e.attention_mask for e in encoded])
+        input_names = {node.name for node in self.model.get_inputs()}  # type: ignore[union-attr]
+        onnx_input: dict[str, NumpyArray] = {
+            "input_ids": np.array(input_ids, dtype=np.int64),
+        }
+        if "attention_mask" in input_names:
+            onnx_input["attention_mask"] = np.array(attention_mask, dtype=np.int64)
+        if "token_type_ids" in input_names:
+            onnx_input["token_type_ids"] = np.array(
+                [np.zeros(len(e), dtype=np.int64) for e in input_ids], dtype=np.int64
+            )
+        onnx_input = self._preprocess_onnx_input(onnx_input, **kwargs)
+
+        model_output = self.model.run(self.ONNX_OUTPUT_NAMES, onnx_input)  # type: ignore[union-attr]
+        return OnnxOutputContext(
+            model_output=model_output[0],
+            attention_mask=onnx_input.get("attention_mask", attention_mask),
+            input_ids=onnx_input.get("input_ids", input_ids),
+        )
+
+    def _embed_documents(
+        self,
+        model_name: str,
+        cache_dir: str,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        **kwargs: Any,
+    ) -> Iterable[T]:
+        is_small = False
+
+        if isinstance(documents, str):
+            documents = [documents]
+            is_small = True
+
+        if isinstance(documents, list):
+            if len(documents) < batch_size:
+                is_small = True
+
+        if parallel is None or is_small:
+            if not hasattr(self, "model") or self.model is None:
+                self.load_onnx_model()
+            for batch in iter_batch(documents, batch_size):
+                yield from self._post_process_onnx_output(self.onnx_embed(batch, **kwargs))
+        else:
+            if parallel == 0:
+                parallel = os.cpu_count()
+
+            start_method = "forkserver" if "forkserver" in get_all_start_methods() else "spawn"
+            params = {
+                "model_name": model_name,
+                "cache_dir": cache_dir,
+                "providers": providers,
+                **kwargs,
+            }
+
+            pool = ParallelWorkerPool(
+                num_workers=parallel or 1,
+                worker=self._get_worker_class(),
+                cuda=cuda,
+                device_ids=device_ids,
+                start_method=start_method,
+            )
+            for batch in pool.ordered_map(iter_batch(documents, batch_size), **params):
+                yield from self._post_process_onnx_output(batch)  # type: ignore
+
+
+class TextEmbeddingWorker(EmbeddingWorker[T]):
+    def process(self, items: Iterable[tuple[int, Any]]) -> Iterable[tuple[int, OnnxOutputContext]]:
+        for idx, batch in items:
+            onnx_output = self.model.onnx_embed(batch)
+            yield idx, onnx_output
@@ -0,0 +1,134 @@
+from typing import Any, Iterable, Type
+
+import numpy as np
+from numpy.typing import NDArray
+
+from fastembed.common.types import NumpyArray
+from fastembed.common.onnx_model import OnnxOutputContext
+from fastembed.common.utils import mean_pooling
+from fastembed.text.onnx_embedding import OnnxTextEmbedding, OnnxTextEmbeddingWorker
+from fastembed.common.model_description import DenseModelDescription, ModelSource
+
+supported_pooled_models: list[DenseModelDescription] = [
+    DenseModelDescription(
+        model="nomic-ai/nomic-embed-text-v1.5",
+        dim=768,
+        description=(
+            "Text embeddings, Multimodal (text, image), English, 8192 input tokens truncation, "
+            "Prefixes for queries/documents: necessary, 2024 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.52,
+        sources=ModelSource(hf="nomic-ai/nomic-embed-text-v1.5"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="nomic-ai/nomic-embed-text-v1.5-Q",
+        dim=768,
+        description=(
+            "Text embeddings, Multimodal (text, image), English, 8192 input tokens truncation, "
+            "Prefixes for queries/documents: necessary, 2024 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.13,
+        sources=ModelSource(hf="nomic-ai/nomic-embed-text-v1.5"),
+        model_file="onnx/model_quantized.onnx",
+    ),
+    DenseModelDescription(
+        model="nomic-ai/nomic-embed-text-v1",
+        dim=768,
+        description=(
+            "Text embeddings, Multimodal (text, image), English, 8192 input tokens truncation, "
+            "Prefixes for queries/documents: necessary, 2024 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.52,
+        sources=ModelSource(hf="nomic-ai/nomic-embed-text-v1"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2",
+        dim=384,
+        description=(
+            "Text embeddings, Unimodal (text), Multilingual (~50 languages), 512 input tokens truncation, "
+            "Prefixes for queries/documents: not necessary, 2019 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.22,
+        sources=ModelSource(hf="qdrant/paraphrase-multilingual-MiniLM-L12-v2-onnx-Q"),
+        model_file="model_optimized.onnx",
+    ),
+    DenseModelDescription(
+        model="sentence-transformers/paraphrase-multilingual-mpnet-base-v2",
+        dim=768,
+        description=(
+            "Text embeddings, Unimodal (text), Multilingual (~50 languages), 384 input tokens truncation, "
+            "Prefixes for queries/documents: not necessary, 2021 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=1.00,
+        sources=ModelSource(hf="xenova/paraphrase-multilingual-mpnet-base-v2"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="intfloat/multilingual-e5-large",
+        dim=1024,
+        description=(
+            "Text embeddings, Unimodal (text), Multilingual (~100 languages), 512 input tokens truncation, "
+            "Prefixes for queries/documents: necessary, 2024 year."
+        ),
+        license="mit",
+        size_in_GB=2.24,
+        sources=ModelSource(
+            hf="qdrant/multilingual-e5-large-onnx",
+            url="https://storage.googleapis.com/qdrant-fastembed/fast-multilingual-e5-large.tar.gz",
+            _deprecated_tar_struct=True,
+        ),
+        model_file="model.onnx",
+        additional_files=["model.onnx_data"],
+    ),
+]
+
+
+class PooledEmbedding(OnnxTextEmbedding):
+    @classmethod
+    def _get_worker_class(cls) -> Type[OnnxTextEmbeddingWorker]:
+        return PooledEmbeddingWorker
+
+    @classmethod
+    def mean_pooling(
+        cls, model_output: NumpyArray, attention_mask: NDArray[np.int64]
+    ) -> NumpyArray:
+        return mean_pooling(model_output, attention_mask)
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        """Lists the supported models.
+
+        Returns:
+            list[DenseModelDescription]: A list of DenseModelDescription objects containing the model information.
+        """
+        return supported_pooled_models
+
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[NumpyArray]:
+        if output.attention_mask is None:
+            raise ValueError("attention_mask must be provided for document post-processing")
+
+        embeddings = output.model_output
+        attn_mask = output.attention_mask
+        return self.mean_pooling(embeddings, attn_mask)
+
+
+class PooledEmbeddingWorker(OnnxTextEmbeddingWorker):
+    def init_embedding(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ) -> OnnxTextEmbedding:
+        return PooledEmbedding(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )
@@ -0,0 +1,162 @@
+from typing import Any, Iterable, Type
+
+
+from fastembed.common.types import NumpyArray
+from fastembed.common.onnx_model import OnnxOutputContext
+from fastembed.common.utils import normalize
+from fastembed.text.onnx_embedding import OnnxTextEmbedding, OnnxTextEmbeddingWorker
+from fastembed.text.pooled_embedding import PooledEmbedding
+from fastembed.common.model_description import DenseModelDescription, ModelSource
+
+supported_pooled_normalized_models: list[DenseModelDescription] = [
+    DenseModelDescription(
+        model="sentence-transformers/all-MiniLM-L6-v2",
+        dim=384,
+        description=(
+            "Text embeddings, Unimodal (text), English, 256 input tokens truncation, "
+            "Prefixes for queries/documents: not necessary, 2021 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.09,
+        sources=ModelSource(
+            url="https://storage.googleapis.com/qdrant-fastembed/sentence-transformers-all-MiniLM-L6-v2.tar.gz",
+            hf="qdrant/all-MiniLM-L6-v2-onnx",
+            _deprecated_tar_struct=True,
+        ),
+        model_file="model.onnx",
+    ),
+    DenseModelDescription(
+        model="jinaai/jina-embeddings-v2-base-en",
+        dim=768,
+        description=(
+            "Text embeddings, Unimodal (text), English, 8192 input tokens truncation, "
+            "Prefixes for queries/documents: not necessary, 2023 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.52,
+        sources=ModelSource(hf="xenova/jina-embeddings-v2-base-en"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="jinaai/jina-embeddings-v2-small-en",
+        dim=512,
+        description=(
+            "Text embeddings, Unimodal (text), English, 8192 input tokens truncation, "
+            "Prefixes for queries/documents: not necessary, 2023 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.12,
+        sources=ModelSource(hf="xenova/jina-embeddings-v2-small-en"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="jinaai/jina-embeddings-v2-base-de",
+        dim=768,
+        description=(
+            "Text embeddings, Unimodal (text), Multilingual (German, English), 8192 input tokens truncation, "
+            "Prefixes for queries/documents: not necessary, 2024 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.32,
+        sources=ModelSource(hf="jinaai/jina-embeddings-v2-base-de"),
+        model_file="onnx/model_fp16.onnx",
+    ),
+    DenseModelDescription(
+        model="jinaai/jina-embeddings-v2-base-code",
+        dim=768,
+        description=(
+            "Text embeddings, Unimodal (text), Multilingual (English, 30 programming languages), "
+            "8192 input tokens truncation, Prefixes for queries/documents: not necessary, 2024 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.64,
+        sources=ModelSource(hf="jinaai/jina-embeddings-v2-base-code"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="jinaai/jina-embeddings-v2-base-zh",
+        dim=768,
+        description=(
+            "Text embeddings, Unimodal (text), supports mixed Chinese-English input text, "
+            "8192 input tokens truncation, Prefixes for queries/documents: not necessary, 2024 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.64,
+        sources=ModelSource(hf="jinaai/jina-embeddings-v2-base-zh"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="jinaai/jina-embeddings-v2-base-es",
+        dim=768,
+        description=(
+            "Text embeddings, Unimodal (text), supports mixed Spanish-English input text, "
+            "8192 input tokens truncation, Prefixes for queries/documents: not necessary, 2024 year."
+        ),
+        license="apache-2.0",
+        size_in_GB=0.64,
+        sources=ModelSource(hf="jinaai/jina-embeddings-v2-base-es"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="thenlper/gte-base",
+        dim=768,
+        description=(
+            "General text embeddings, Unimodal (text), supports English only input text, "
+            "512 input tokens truncation, Prefixes for queries/documents: not necessary, 2024 year."
+        ),
+        license="mit",
+        size_in_GB=0.44,
+        sources=ModelSource(hf="thenlper/gte-base"),
+        model_file="onnx/model.onnx",
+    ),
+    DenseModelDescription(
+        model="thenlper/gte-large",
+        dim=1024,
+        description=(
+            "Text embeddings, Unimodal (text), English, 512 input tokens truncation, "
+            "Prefixes for queries/documents: not necessary, 2023 year."
+        ),
+        license="mit",
+        size_in_GB=1.20,
+        sources=ModelSource(hf="qdrant/gte-large-onnx"),
+        model_file="model.onnx",
+    ),
+]
+
+
+class PooledNormalizedEmbedding(PooledEmbedding):
+    @classmethod
+    def _get_worker_class(cls) -> Type[OnnxTextEmbeddingWorker]:
+        return PooledNormalizedEmbeddingWorker
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        """Lists the supported models.
+
+        Returns:
+            list[DenseModelDescription]: A list of DenseModelDescription objects containing the model information.
+        """
+        return supported_pooled_normalized_models
+
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[NumpyArray]:
+        if output.attention_mask is None:
+            raise ValueError("attention_mask must be provided for document post-processing")
+
+        embeddings = output.model_output
+        attn_mask = output.attention_mask
+        return normalize(self.mean_pooling(embeddings, attn_mask))
+
+
+class PooledNormalizedEmbeddingWorker(OnnxTextEmbeddingWorker):
+    def init_embedding(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
+    ) -> OnnxTextEmbedding:
+        return PooledNormalizedEmbedding(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )
@@ -0,0 +1,180 @@
+import warnings
+from typing import Any, Iterable, Optional, Sequence, Type, Union
+from dataclasses import asdict
+
+from fastembed.common.types import NumpyArray, OnnxProvider
+from fastembed.text.clip_embedding import CLIPOnnxEmbedding
+from fastembed.text.custom_text_embedding import CustomTextEmbedding
+from fastembed.text.pooled_normalized_embedding import PooledNormalizedEmbedding
+from fastembed.text.pooled_embedding import PooledEmbedding
+from fastembed.text.multitask_embedding import JinaEmbeddingV3
+from fastembed.text.onnx_embedding import OnnxTextEmbedding
+from fastembed.text.text_embedding_base import TextEmbeddingBase
+from fastembed.common.model_description import DenseModelDescription, ModelSource, PoolingType
+
+
+class TextEmbedding(TextEmbeddingBase):
+    EMBEDDINGS_REGISTRY: list[Type[TextEmbeddingBase]] = [
+        OnnxTextEmbedding,
+        CLIPOnnxEmbedding,
+        PooledNormalizedEmbedding,
+        PooledEmbedding,
+        JinaEmbeddingV3,
+        CustomTextEmbedding,
+    ]
+
+    @classmethod
+    def list_supported_models(cls) -> list[dict[str, Any]]:
+        """Lists the supported models.
+
+        Returns:
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
+        """
+        return [asdict(model) for model in cls._list_supported_models()]
+
+    @classmethod
+    def _list_supported_models(cls) -> list[DenseModelDescription]:
+        result: list[DenseModelDescription] = []
+        for embedding in cls.EMBEDDINGS_REGISTRY:
+            result.extend(embedding._list_supported_models())
+        return result
+
+    @classmethod
+    def add_custom_model(
+        cls,
+        model: str,
+        pooling: PoolingType,
+        normalization: bool,
+        sources: ModelSource,
+        dim: int,
+        model_file: str = "onnx/model.onnx",
+        description: str = "",
+        license: str = "",
+        size_in_gb: float = 0.0,
+        additional_files: Optional[list[str]] = None,
+    ) -> None:
+        registered_models = cls._list_supported_models()
+        for registered_model in registered_models:
+            if model == registered_model.model:
+                raise ValueError(
+                    f"Model {model} is already registered in TextEmbedding, if you still want to add this model, "
+                    f"please use another model name"
+                )
+
+        CustomTextEmbedding.add_model(
+            DenseModelDescription(
+                model=model,
+                sources=sources,
+                dim=dim,
+                model_file=model_file,
+                description=description,
+                license=license,
+                size_in_GB=size_in_gb,
+                additional_files=additional_files or [],
+            ),
+            pooling=pooling,
+            normalization=normalization,
+        )
+
+    def __init__(
+        self,
+        model_name: str = "BAAI/bge-small-en-v1.5",
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        lazy_load: bool = False,
+        **kwargs: Any,
+    ):
+        super().__init__(model_name, cache_dir, threads, **kwargs)
+        if model_name == "nomic-ai/nomic-embed-text-v1.5-Q":
+            warnings.warn(
+                "The model 'nomic-ai/nomic-embed-text-v1.5-Q' has been updated on HuggingFace. Please review "
+                "the latest documentation on HF and release notes to ensure compatibility with your workflow. ",
+                UserWarning,
+                stacklevel=2,
+            )
+        if model_name in {
+            "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2",
+            "thenlper/gte-large",
+            "intfloat/multilingual-e5-large",
+            "sentence-transformers/paraphrase-multilingual-mpnet-base-v2",
+        }:
+            warnings.warn(
+                f"The model {model_name} now uses mean pooling instead of CLS embedding. "
+                f"In order to preserve the previous behaviour, consider either pinning fastembed version to 0.5.1 or "
+                "using `add_custom_model` functionality.",
+                UserWarning,
+                stacklevel=2,
+            )
+        for EMBEDDING_MODEL_TYPE in self.EMBEDDINGS_REGISTRY:
+            supported_models = EMBEDDING_MODEL_TYPE._list_supported_models()
+            if any(model_name.lower() == model.model.lower() for model in supported_models):
+                self.model = EMBEDDING_MODEL_TYPE(
+                    model_name=model_name,
+                    cache_dir=cache_dir,
+                    threads=threads,
+                    providers=providers,
+                    cuda=cuda,
+                    device_ids=device_ids,
+                    lazy_load=lazy_load,
+                    **kwargs,
+                )
+                return
+
+        raise ValueError(
+            f"Model {model_name} is not supported in TextEmbedding. "
+            "Please check the supported models using `TextEmbedding.list_supported_models()`"
+        )
+
+    def embed(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        """
+        Encode a list of documents into list of embeddings.
+        We use mean pooling with attention so that the model can handle variable-length inputs.
+
+        Args:
+            documents: Iterator of documents or single document to embed
+            batch_size: Batch size for encoding -- higher values will use more memory, but be faster
+            parallel:
+                If > 1, data-parallel encoding will be used, recommended for offline encoding of large datasets.
+                If 0, use all available cores.
+                If None, don't use data-parallel processing, use default onnxruntime threading instead.
+
+        Returns:
+            List of embeddings, one per document
+        """
+        yield from self.model.embed(documents, batch_size, parallel, **kwargs)
+
+    def query_embed(self, query: Union[str, Iterable[str]], **kwargs: Any) -> Iterable[NumpyArray]:
+        """
+        Embeds queries
+
+        Args:
+            query (Union[str, Iterable[str]]): The query to embed, or an iterable e.g. list of queries.
+
+        Returns:
+            Iterable[NumpyArray]: The embeddings.
+        """
+        # This is model-specific, so that different models can have specialized implementations
+        yield from self.model.query_embed(query, **kwargs)
+
+    def passage_embed(self, texts: Iterable[str], **kwargs: Any) -> Iterable[NumpyArray]:
+        """
+        Embeds a list of text passages into a list of embeddings.
+
+        Args:
+            texts (Iterable[str]): The list of texts to embed.
+            **kwargs: Additional keyword argument to pass to the embed method.
+
+        Yields:
+            Iterable[SparseEmbedding]: The sparse embeddings.
+        """
+        # This is model-specific, so that different models can have specialized implementations
+        yield from self.model.passage_embed(texts, **kwargs)
@@ -0,0 +1,60 @@
+from typing import Iterable, Optional, Union, Any
+
+from fastembed.common.model_description import DenseModelDescription
+from fastembed.common.types import NumpyArray
+from fastembed.common.model_management import ModelManagement
+
+
+class TextEmbeddingBase(ModelManagement[DenseModelDescription]):
+    def __init__(
+        self,
+        model_name: str,
+        cache_dir: Optional[str] = None,
+        threads: Optional[int] = None,
+        **kwargs: Any,
+    ):
+        self.model_name = model_name
+        self.cache_dir = cache_dir
+        self.threads = threads
+        self._local_files_only = kwargs.pop("local_files_only", False)
+
+    def embed(
+        self,
+        documents: Union[str, Iterable[str]],
+        batch_size: int = 256,
+        parallel: Optional[int] = None,
+        **kwargs: Any,
+    ) -> Iterable[NumpyArray]:
+        raise NotImplementedError()
+
+    def passage_embed(self, texts: Iterable[str], **kwargs: Any) -> Iterable[NumpyArray]:
+        """
+        Embeds a list of text passages into a list of embeddings.
+
+        Args:
+            texts (Iterable[str]): The list of texts to embed.
+            **kwargs: Additional keyword argument to pass to the embed method.
+
+        Yields:
+            Iterable[NumpyArray]: The embeddings.
+        """
+
+        # This is model-specific, so that different models can have specialized implementations
+        yield from self.embed(texts, **kwargs)
+
+    def query_embed(self, query: Union[str, Iterable[str]], **kwargs: Any) -> Iterable[NumpyArray]:
+        """
+        Embeds queries
+
+        Args:
+            query (Union[str, Iterable[str]]): The query to embed, or an iterable e.g. list of queries.
+
+        Returns:
+            Iterable[NumpyArray]: The embeddings.
+        """
+
+        # This is model-specific, so that different models can have specialized implementations
+        if isinstance(query, str):
+            yield from self.embed([query], **kwargs)
+        else:
+            yield from self.embed(query, **kwargs)
@@ -2,7 +2,7 @@ site_name: FastEmbed
 site_url: https://qdrant.github.io/fastembed/
 site_author: Nirant Kasliwal
 repo_url: https://github.com/qdrant/fastembed/
-repo_name: qdrant/fastembed  
+repo_name: qdrant/fastembed

 remote_branch: gh-pages
 remote_name: origin
@@ -33,11 +33,11 @@ theme:
    # Text color for primary color
    text: "#ffffff"

-  palette: 
+  palette:
    # Palette toggle for light mode
    - scheme: default
      toggle:
-        icon: material/brightness-7 
+        icon: material/brightness-7
        name: Switch to dark mode

    # Palette toggle for dark mode
@@ -1,8 +1,8 @@
 [tool.poetry]
-name = "fastembed"
-version = "0.0.4"
+name = "fastembed-gpu"
+version = "0.6.1"
 description = "Fast, light, accurate library built for retrieval embedding generation"
-authors = ["NirantK <nirant.bits@gmail.com>"]
+authors = ["Qdrant Team <info@qdrant.tech>", "NirantK <nirant.bits@gmail.com>"]
 license = "Apache License"
 readme = "README.md"
 packages = [{include = "fastembed"}]
@@ -11,40 +11,53 @@ repository = "https://github.com/qdrant/fastembed"
 keywords = ["vector", "embedding", "neural", "search", "qdrant", "sentence-transformers"]

 [tool.poetry.dependencies]
-python = ">=3.8.0,<3.12"
-onnxruntime = "^1.15.1"
-torch = ">=2.0.0, !=2.0.1"
-optimum = ">1.12.0"
-tqdm = "^4.65.0"
-requests = "^2.31.0"
-tokenizers = "^0.13.3"
+python = ">=3.9.0"
+numpy = [
+    { version = ">=1.21", python = ">=3.10,<3.12" },
+    { version = ">=1.26", python = ">=3.12,<3.13" },
+    { version = ">=2.1.0", python = ">=3.13" },
+    { version = ">=1.21,<2.1.0", python = "<3.10" },
+]
+onnxruntime-gpu = [
+    { version = ">1.20.0", python = ">=3.13" },
+    { version = ">=1.17.0,<1.20.0", python = "<3.10" },
+    { version = ">=1.17.0,!=1.20.0", python = ">=3.10,<3.13" },
+]
+tqdm = "^4.66"
+requests = "^2.31"
+tokenizers = ">=0.15,<1.0"
+huggingface-hub = ">=0.20,<1.0"
+loguru = "^0.7.2"
+pillow = ">=10.3.0,<12.0.0"
+mmh3 = ">=4.1.0,<6.0.0"
+py-rust-stemmers = "^0.1.0"

-[tool.poetry.dev-dependencies]
-ruff = "^0.0.277"
-isort = "^5.12.0"
-black = "^23.7.0"
-onnx = "^1.11.0"
+[tool.poetry.group.test.dependencies]
+pytest = "^7.4.2"
+ruff = ">=0.3.1,<1.0"
+
+[tool.poetry.group.dev.dependencies]
 notebook = ">=7.0.2"
-mkdocs-material = "^9.1.21"
-mkdocstrings = "^0.22.0"
-pillow = "^10.0.0"
+pre-commit = "^3.6.2"
+onnx = ">=1.15.0"
+
+[tool.poetry.group.docs.dependencies]
+mkdocs-material = "^9.5.10"
+mkdocstrings = "^0.24.0"
+pillow = ">=10.3.0,<12.0.0"
 cairosvg = "^2.7.1"
 mknotebooks = "^0.8.0"
-pytest = "^7.4.0"

-[tool.poetry.dependencies.onnxruntime-silicon]
-version = "^1.15.0"
-markers = "sys_platform == 'darwin'"  # This makes it macOS specific
+[tool.poetry.group.types.dependencies]
+pyright = ">=1.1.293"
+mypy = "^1.0.0"

 [build-system]
 requires = ["poetry-core"]
 build-backend = "poetry.core.masonry.api"

-[tool.black]
-line-length = 120
-
-[tool.isort]
-profile = "black"
+[tool.pyright]
+typeCheckingMode = "strict"

 [tool.ruff]
-line-length = 120
+line-length = 99
@@ -0,0 +1,4 @@
+import os
+
+# disable DeprecationWarning https://github.com/jupyter/jupyter_core/issues/398
+os.environ["JUPYTER_PLATFORM_DIRS"] = "1"
@@ -0,0 +1,4 @@
+from pathlib import Path
+
+TEST_DIR = Path(__file__).parent
+TEST_MISC_DIR = TEST_DIR / "misc"
@@ -1,31 +1,29 @@
 # %% [markdown]
 # # 🤗 Huggingface vs ⚡ FastEmbed️
-# 
+#
 # Comparing the performance of Huggingface's 🤗 Transformers and ⚡ FastEmbed️ on a simple task on the following machine: Apple M2 Max, 32 GB RAM
-# 
+#
 # ## 📦 Imports
-# 
+#
 # Importing the necessary libraries for this comparison.

 # %%
 import time
-from pathlib import Path
-from typing import Any, Callable, List, Tuple
+from typing import Callable

-import numpy as np
+import matplotlib.pyplot as plt
 import torch.nn.functional as F
-from torch import Tensor
 from transformers import AutoModel, AutoTokenizer

 from fastembed.embedding import DefaultEmbedding

 # %% [markdown]
 # ## 📖 Data
-# 
+#
 # data is a list of strings, each string is a document.

 # %%
-documents: List[str] = [
+documents: list[str] = [
    "Chandrayaan-3 is India's third lunar mission",
    "It aimed to land a rover on the Moon's surface - joining the US, China and Russia",
    "The mission is a follow-up to Chandrayaan-2, which had partial success",
@@ -43,9 +41,10 @@ len(documents)

 # %% [markdown]
 # ## Setting up 🤗 Huggingface
-# 
+#
 # We'll be using the [Huggingface Transformers](https://huggingface.co/transformers/) with PyTorch library to generate embeddings. We'll be using the same model across both libraries for a fair(er?) comparison.

+
 # %%
 class HF:
    """
@@ -57,19 +56,22 @@ class HF:
        self.model = AutoModel.from_pretrained(model_id)
        self.tokenizer = AutoTokenizer.from_pretrained(model_id)

-    def embed(self, texts: List[str]):
-        encoded_input = self.tokenizer(texts, max_length=512, padding=True, truncation=True, return_tensors="pt")
+    def embed(self, texts: list[str]):
+        encoded_input = self.tokenizer(
+            texts, max_length=512, padding=True, truncation=True, return_tensors="pt"
+        )
        model_output = self.model(**encoded_input)
        sentence_embeddings = model_output[0][:, 0]
        sentence_embeddings = F.normalize(sentence_embeddings)
        return sentence_embeddings

+
 hf = HF(model_id="BAAI/bge-small-en")
 hf.embed(documents).shape

 # %% [markdown]
 # ## Setting up ⚡️FastEmbed
-# 
+#
 # Sorry, don't have a lot to set up here. We'll be using the default model, which is Flag Embedding, same as the Huggingface model.

 # %%
@@ -77,18 +79,21 @@ embedding_model = DefaultEmbedding()

 # %% [markdown]
 # ## 📊 Comparison
-# 
+#
 # We'll be comparing the following metrics: Minimum, Maximum, Mean, across k runs. Let's write a function to do that:
-# 
+#
 # ### 🚀 Calculating Stats

+
 # %%
-def calculate_time_stats(embed_func: Callable, documents: list, k: int) -> Tuple[float, float, float]:
+def calculate_time_stats(
+    embed_func: Callable, documents: list, k: int
+) -> tuple[float, float, float]:
    times = []
    for _ in range(k):
        # Timing the embed_func call
        start_time = time.time()
-        embeddings = embed_func(documents)
+        embed_func(documents)
        end_time = time.time()

        times.append(end_time - start_time)
@@ -96,19 +101,19 @@ def calculate_time_stats(embed_func: Callable, documents: list, k: int) -> Tuple
    # Returning mean, max, and min time for the call
    return (sum(times) / k, max(times), min(times))

+
 # %%
 hf_stats = calculate_time_stats(hf.embed, documents, k=2)
 print(f"Huggingface Transformers (Average, Max, Min): {hf_stats}")
 fst_stats = calculate_time_stats(lambda x: list(embedding_model.embed(x)), documents, k=2)
 print(f"FastEmbed (Average, Max, Min): {fst_stats}")

-# %%
-import matplotlib.pyplot as plt
-

 # %%
 def plot_character_per_second_comparison(
-    hf_stats: Tuple[float, float, float], fst_stats: Tuple[float, float, float], documents: list
+    hf_stats: tuple[float, float, float],
+    fst_stats: tuple[float, float, float],
+    documents: list,
 ):
    # Calculating total characters in documents
    total_characters = sum(len(doc) for doc in documents)
@@ -141,5 +146,3 @@ def plot_character_per_second_comparison(


 plot_character_per_second_comparison(hf_stats, fst_stats, documents)
-
-
@@ -0,0 +1,159 @@
+import os
+
+import numpy as np
+import pytest
+
+from fastembed import SparseTextEmbedding
+from tests.utils import delete_model_cache
+
+
+@pytest.mark.parametrize("model_name", ["Qdrant/bm42-all-minilm-l6-v2-attentions", "Qdrant/bm25"])
+def test_attention_embeddings(model_name: str) -> None:
+    is_ci = os.getenv("CI")
+    model = SparseTextEmbedding(model_name=model_name)
+
+    output = list(
+        model.query_embed(
+            [
+                "I must not fear. Fear is the mind-killer.",
+            ]
+        )
+    )
+
+    assert len(output) == 1
+
+    for result in output:
+        assert len(result.indices) == len(result.values)
+        assert np.allclose(result.values, np.ones(len(result.values)))
+
+    quotes = [
+        "I must not fear. Fear is the mind-killer.",
+        "All animals are equal, but some animals are more equal than others.",
+        "It was a pleasure to burn.",
+        "The sky above the port was the color of television, tuned to a dead channel.",
+        "In the beginning, the universe was created."
+        " This has made a lot of people very angry and been widely regarded as a bad move.",
+        "It's a truth universally acknowledged that a zombie in possession of brains must be in want of more brains.",
+        "War is peace. Freedom is slavery. Ignorance is strength.",
+        "We're not in Infinity; we're in the suburbs.",
+        "I was a thousand times more evil than thou!",
+        "History is merely a list of surprises... It can only prepare us to be surprised yet again.",
+        ".",  # Empty string
+    ]
+
+    output = list(model.embed(quotes))
+
+    assert len(output) == len(quotes)
+
+    for result in output[:-1]:
+        assert len(result.indices) == len(result.values)
+        assert len(result.indices) > 0
+
+    assert len(output[-1].indices) == 0
+
+    # Test support for unknown languages
+    output = list(
+        model.query_embed(
+            [
+                "привет мир!",
+            ]
+        )
+    )
+
+    assert len(output) == 1
+
+    for result in output:
+        assert len(result.indices) == len(result.values)
+        assert len(result.indices) == 2
+
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
+
+
+@pytest.mark.parametrize("model_name", ["Qdrant/bm42-all-minilm-l6-v2-attentions", "Qdrant/bm25"])
+def test_parallel_processing(model_name: str) -> None:
+    is_ci = os.getenv("CI")
+
+    model = SparseTextEmbedding(model_name=model_name)
+
+    docs = ["hello world", "attention embedding", "Mangez-vous vraiment des grenouilles?"] * 100
+    embeddings = list(model.embed(docs, batch_size=10, parallel=2))
+
+    embeddings_2 = list(model.embed(docs, batch_size=10, parallel=None))
+
+    embeddings_3 = list(model.embed(docs, batch_size=10, parallel=0))
+
+    assert len(embeddings) == len(docs)
+
+    for emb_1, emb_2, emb_3 in zip(embeddings, embeddings_2, embeddings_3):
+        assert np.allclose(emb_1.indices, emb_2.indices)
+        assert np.allclose(emb_1.indices, emb_3.indices)
+        assert np.allclose(emb_1.values, emb_2.values)
+        assert np.allclose(emb_1.values, emb_3.values)
+
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
+
+
+@pytest.mark.parametrize("model_name", ["Qdrant/bm25"])
+def test_multilanguage(model_name: str) -> None:
+    is_ci = os.getenv("CI")
+
+    docs = ["Mangez-vous vraiment des grenouilles?", "Je suis au lit"]
+
+    model = SparseTextEmbedding(model_name=model_name, language="french")
+    embeddings = list(model.embed(docs))[:2]
+    assert embeddings[0].values.shape == (3,)
+    assert embeddings[0].indices.shape == (3,)
+
+    assert embeddings[1].values.shape == (1,)
+    assert embeddings[1].indices.shape == (1,)
+
+    model = SparseTextEmbedding(model_name=model_name, language="english")
+    embeddings = list(model.embed(docs))[:2]
+    assert embeddings[0].values.shape == (5,)
+    assert embeddings[0].indices.shape == (5,)
+
+    assert embeddings[1].values.shape == (4,)
+    assert embeddings[1].indices.shape == (4,)
+
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
+
+
+@pytest.mark.parametrize("model_name", ["Qdrant/bm25"])
+def test_special_characters(model_name: str) -> None:
+    is_ci = os.getenv("CI")
+
+    docs = [
+        "Über den größten Flüssen Österreichs äußern sich Experten häufig: Öko-Systeme müssen geschützt werden!",
+        "L'élève français s'écrie : « Où est mon crayon ? J'ai besoin de finir cet exercice avant la récréation!",
+        "Într-o zi însorită, Ștefan și Ioana au mâncat mămăligă cu brânză și au băut țuică la cabană.",
+        "Üzgün öğretmen öğrencilere seslendi: Lütfen gürültü yapmayın, sınavınızı bitirmeye çalışıyorum!",
+        "Ο Ξενοφών είπε: «Ψάχνω για ένα ωραίο δώρο για τη γιαγιά μου. Ίσως ένα φυτό ή ένα βιβλίο;»",
+        "Hola! ¿Cómo estás? Estoy muy emocionado por el cumpleaños de mi hermano, ¡va a ser increíble! También quiero comprar un pastel de chocolate con fresas y un regalo especial: un libro titulado «Cien años de soledad",
+    ]
+
+    model = SparseTextEmbedding(model_name=model_name, language="english")
+    embeddings = list(model.embed(docs))
+    for idx, shape in enumerate([14, 18, 15, 10, 15]):
+        assert embeddings[idx].values.shape == (shape,)
+        assert embeddings[idx].indices.shape == (shape,)
+
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
+
+
+@pytest.mark.parametrize("model_name", ["Qdrant/bm42-all-minilm-l6-v2-attentions"])
+def test_lazy_load(model_name: str) -> None:
+    model = SparseTextEmbedding(model_name=model_name, lazy_load=True)
+    assert not hasattr(model.model, "model")
+    docs = ["hello world", "flag embedding"]
+    list(model.embed(docs))
+    assert hasattr(model.model, "model")
+
+    model = SparseTextEmbedding(model_name=model_name, lazy_load=True)
+    list(model.query_embed(docs))
+
+    model = SparseTextEmbedding(model_name=model_name, lazy_load=True)
+    list(model.passage_embed(docs))
--- a/Show More
+++ b/Show More