Merge branch 'main' into mean_pooling

* chore: Add missing type hints in functions

* add missing import, small type refactor

---------

Co-authored-by: George Panchuk <george.panchuk@qdrant.tech>

.github/ISSUE_TEMPLATE/bug-report.yml

@@ -1,6 +1,6 @@
-name: Bug/New Model Request
-description: File a bug report/Request a new Model
-title: "[Bug/Model Request]: "
+name: Bug
+description: File a bug report
+title: "[Bug]: "
 body:
   - type: markdown
     attributes:
@@ -10,11 +10,22 @@ body:
     id: what-happened
     attributes:
       label: What happened?
-      description: Also tell us, what did you expect to happen?
-      placeholder: Tell us what you see!
-      value: "A bug 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:
@@ -23,21 +34,12 @@ body:
       placeholder: Python3.10
     validations:
       required: true
-  - type: dropdown
+  - type: textarea
     id: version
     attributes:
-      label: Version
+      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.
-      options:
-        - 0.2.7 (Latest)
-        - 0.2.6
-        - 0.2.5
-        - 0.2.4 
-        - 0.2.3 
-        - 0.2.2
-        - 0.2.1
-        - 0.1.x 
-      default: 0
+      placeholder: v0.5.1
     validations:
       required: true
   - type: dropdown

.github/ISSUE_TEMPLATE/config.yaml

@@ -1,4 +1,4 @@
-blank_issues_enabled: false
+blank_issues_enabled: true
 contact_links:
   - name: GitHub Community Support
     url: https://github.com/qdrant/fastembed/discussions

.github/ISSUE_TEMPLATE/feature-request.yml

@@ -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>

.github/ISSUE_TEMPLATE/model-request.yml

@@ -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>

.github/PULL_REQUEST_TEMPLATE.md

@@ -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?

.github/workflows/python-tests.yml

@@ -3,8 +3,6 @@ name: Tests
 on:
   push:
     branches: [ master, main, gpu ]
-  schedule:
-    - cron: 0 0 * * *
   pull_request:
 
 env:
@@ -16,11 +14,11 @@ 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
           - macos-latest
@@ -40,15 +38,8 @@ jobs:
         run: |
           python -m pip install poetry
           poetry config virtualenvs.create false
-          poetry install --no-interaction --no-ansi --without docs
-
-      - name: Install Test Dependencies
-        run: pip install pytest pytest-md pytest-emoji
+          poetry install --no-interaction --no-ansi --without dev,docs
 
       - name: Run pytest
-        uses: pavelzw/pytest-action@v2
-        with:
-            verbose: true
-            emoji: true
-            job-summary: true
-            report-title: 'FastEmbed Test Report'
+        run: |
+          poetry run pytest 

NOTICE

@@ -0,0 +1,16 @@
+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.
+
+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.

README.md

@@ -28,10 +28,10 @@ pip install fastembed-gpu
 
 ```python
 from fastembed import TextEmbedding
-from typing import List
+
 
 # Example list of documents
-documents: List[str] = [
+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.",
 ]
@@ -137,6 +137,20 @@ embeddings = list(model.embed(images))
 # ]
 ```
 
+### 🔄 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]
+```
 
 ## ⚡️ FastEmbed on a GPU
 
@@ -147,7 +161,7 @@ It requires installation of the `fastembed-gpu` package.
 pip install fastembed-gpu
 ```
 
-Check our [example](https://qdrant.github.io/fastembed/examples/FastEmbed_GPU/) for detailed instructions and CUDA 12.x support.
+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

docs/Getting Started.ipynb

@@ -65,15 +65,13 @@
     }
    ],
    "source": [
-    "from typing import List\n",
-    "\n",
     "import numpy as np\n",
     "\n",
     "from fastembed import TextEmbedding\n",
     "\n",
     "\n",
     "# Example list of documents\n",
-    "documents: List[str] = [\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",

docs/examples/ColBERT_with_FastEmbed.ipynb

@@ -54,7 +54,14 @@
     },
     {
      "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'}]"
+      "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": {},
@@ -212,7 +219,9 @@
    "outputs": [
     {
      "data": {
-      "text/plain": "((26, 128), (32, 128))"
+      "text/plain": [
+       "((26, 128), (32, 128))"
+      ]
      },
      "execution_count": 18,
      "metadata": {},
@@ -271,7 +280,9 @@
     "import numpy as np\n",
     "\n",
     "\n",
-    "def compute_relevance_scores(query_embedding: np.array, document_embeddings: np.array, k: int):\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",

docs/examples/FastEmbed_GPU.ipynb

@@ -37,28 +37,7 @@
     "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`.\n",
-    "\n",
-    "### CUDA 12.x support\n",
-    "\n",
-    "By default `onnxruntime-gpu` is shipped with CUDA 11.8 support.\n",
-    "CUDA 12.x support requires installation of `onnxruntime-gpu` with providing of a direct url:"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {
-    "colab": {
-     "base_uri": "https://localhost:8080/"
-    },
-    "id": "OoSfWFFZeJ5t",
-    "outputId": "417b9332-6a7b-4000-c74b-4ed2b5b76590"
-   },
-   "outputs": [],
-   "source": [
-    "!pip install onnxruntime-gpu -i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/simple/ -qq\n",
-    "!pip install fastembed-gpu -qqq"
+    "**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`."
    ]
   },
   {
@@ -67,9 +46,65 @@
     "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",
-    "Google Colab notebooks have CUDA 12.x."
+    "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."
    ]
   },
   {
@@ -82,7 +117,80 @@
     "\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",
-    "Dependencies required for the chosen onnxruntime version can be found [here](https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements)"
+    "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",
+    "```"
    ]
   },
   {
@@ -280,8 +388,6 @@
     }
    ],
    "source": [
-    "from typing import List\n",
-    "\n",
     "import numpy as np\n",
     "\n",
     "from fastembed import TextEmbedding\n",
@@ -299,9 +405,7 @@
     "id": "iPtoHf7GeV-i"
    },
    "outputs": [],
-   "source": [
-    "documents: List[str] = list(np.repeat(\"Demonstrating GPU acceleration in fastembed\", 500))"
-   ]
+   "source": "documents: list[str] = list(np.repeat(\"Demonstrating GPU acceleration in fastembed\", 500))"
   },
   {
    "cell_type": "code",

docs/examples/FastEmbed_Multi_GPU.ipynb

@@ -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
+}

docs/examples/FastEmbed_vs_HF_Comparison.ipynb

@@ -36,7 +36,7 @@
    "outputs": [],
    "source": [
     "import time\n",
-    "from typing import Callable, List, Tuple\n",
+    "from typing import Callable\n",
     "\n",
     "import matplotlib.pyplot as plt\n",
     "import torch.nn.functional as F\n",
@@ -64,6 +64,7 @@
    ],
    "source": [
     "import fastembed\n",
+    "\n",
     "fastembed.__version__"
    ]
   },
@@ -98,7 +99,7 @@
     }
    ],
    "source": [
-    "documents: List[str] = [\n",
+    "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",
@@ -151,11 +152,11 @@
     "    HuggingFace Transformer implementation of FlagEmbedding\n",
     "    \"\"\"\n",
     "\n",
-    "    def __init__(self, model_id: str):\n",
+    "    def __init__(self, model_id: str) -> None:\n",
     "        self.model = AutoModel.from_pretrained(model_id)\n",
     "        self.tokenizer = AutoTokenizer.from_pretrained(model_id)\n",
     "\n",
-    "    def embed(self, texts: List[str]):\n",
+    "    def embed(self, texts: list[str]):\n",
     "        encoded_input = self.tokenizer(\n",
     "            texts, max_length=512, padding=True, truncation=True, return_tensors=\"pt\"\n",
     "        )\n",
@@ -254,7 +255,7 @@
     "\n",
     "def calculate_time_stats(\n",
     "    embed_func: Callable, documents: list, k: int\n",
-    ") -> Tuple[float, float, float]:\n",
+    ") -> tuple[float, float, float]:\n",
     "    times = []\n",
     "    for _ in range(k):\n",
     "        # Timing the embed_func call\n",
@@ -309,7 +310,7 @@
    ],
    "source": [
     "def plot_character_per_second_comparison(\n",
-    "    hf_stats: Tuple[float, float, float], fst_stats: Tuple[float, float, float], documents: list\n",
+    "    hf_stats: tuple[float, float, float], fst_stats: tuple[float, float, float], documents: list\n",
     "):\n",
     "    # Calculating total characters in documents\n",
     "    total_characters = sum(len(doc) for doc in documents)\n",

docs/examples/Hindi_Tamil_RAG_with_Navarasa7B.ipynb

@@ -44,7 +44,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 21,
+   "execution_count": null,
    "metadata": {
     "ExecuteTime": {
      "end_time": "2024-03-30T00:45:24.814968Z",
@@ -58,8 +58,6 @@
    },
    "outputs": [],
    "source": [
-    "from typing import List\n",
-    "\n",
     "import numpy as np\n",
     "from datasets import load_dataset\n",
     "from peft import AutoPeftModelForCausalLM\n",
@@ -72,11 +70,11 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 23,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
-    "hf_token = <YOUR_HF_TOKEN_HERE>  # Get your token from https://huggingface.co/settings/token, needed for Gemma weights"
+    "hf_token = \"<YOUR_HF_TOKEN_HERE>\"  # Get your token from https://huggingface.co/settings/token, needed for Gemma weights"
    ]
   },
   {
@@ -246,7 +244,7 @@
    },
    "outputs": [],
    "source": [
-    "context_embeddings: List[np.ndarray] = list(\n",
+    "context_embeddings: list[np.ndarray] = list(\n",
     "    embedding_model.embed(contexts)\n",
     ")  # Note the list() call - this is a generator"
    ]

docs/examples/Hybrid_Search.ipynb

@@ -50,7 +50,6 @@
    "outputs": [],
    "source": [
     "import json\n",
-    "from typing import List, Tuple\n",
     "\n",
     "import numpy as np\n",
     "import pandas as pd\n",
@@ -489,11 +488,11 @@
     }
    ],
    "source": [
-    "def make_sparse_embedding(texts: List[str]):\n",
+    "def make_sparse_embedding(texts: list[str]) -> list[SparseEmbedding]:\n",
     "    return list(sparse_model.embed(texts, batch_size=32))\n",
     "\n",
     "\n",
-    "sparse_embedding: List[SparseEmbedding] = make_sparse_embedding(\n",
+    "sparse_embedding: list[SparseEmbedding] = make_sparse_embedding(\n",
     "    [\"Fastembed is a great library for text embeddings!\"]\n",
     ")\n",
     "sparse_embedding"
@@ -616,7 +615,7 @@
     }
    ],
    "source": [
-    "def get_tokens_and_weights(sparse_embedding, model_name):\n",
+    "def get_tokens_and_weights(sparse_embedding, model_name) -> dict[str, float]:\n",
     "    # Find the tokenizer for the model\n",
     "    tokenizer_source = None\n",
     "    for model_info in SparseTextEmbedding.list_supported_models():\n",
@@ -627,7 +626,7 @@
     "            raise ValueError(f\"Model {model_name} not found in the supported models.\")\n",
     "\n",
     "    tokenizer = AutoTokenizer.from_pretrained(tokenizer_source)\n",
-    "    token_weight_dict = {}\n",
+    "    token_weight_dict: dict[str, float] = {}\n",
     "    for i in range(len(sparse_embedding.indices)):\n",
     "        token = tokenizer.decode([sparse_embedding.indices[i]])\n",
     "        weight = sparse_embedding.values[i]\n",
@@ -662,7 +661,7 @@
    },
    "outputs": [],
    "source": [
-    "def make_dense_embedding(texts: List[str]):\n",
+    "def make_dense_embedding(texts: list[str]):\n",
     "    return list(dense_model.embed(texts))\n",
     "\n",
     "\n",
@@ -872,7 +871,7 @@
    },
    "outputs": [],
    "source": [
-    "def make_points(df: pd.DataFrame) -> List[PointStruct]:\n",
+    "def make_points(df: pd.DataFrame) -> list[PointStruct]:\n",
     "    sparse_vectors = df[\"sparse_embedding\"].tolist()\n",
     "    product_texts = df[\"combined_text\"].tolist()\n",
     "    dense_vectors = df[\"dense_embedding\"].tolist()\n",
@@ -899,7 +898,7 @@
     "    return points\n",
     "\n",
     "\n",
-    "points: List[PointStruct] = make_points(df)"
+    "points: list[PointStruct] = make_points(df)"
    ]
   },
   {
@@ -942,8 +941,8 @@
    "source": [
     "def search(query_text: str):\n",
     "    # # Compute sparse and dense vectors\n",
-    "    query_sparse_vectors: List[SparseEmbedding] = make_sparse_embedding([query_text])\n",
-    "    query_dense_vector: List[np.ndarray] = make_dense_embedding([query_text])\n",
+    "    query_sparse_vectors: list[SparseEmbedding] = make_sparse_embedding([query_text])\n",
+    "    query_dense_vector: list[np.ndarray] = make_dense_embedding([query_text])\n",
     "\n",
     "    search_results = client.search_batch(\n",
     "        collection_name=collection_name,\n",
@@ -1075,7 +1074,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "def rank_list(search_result: List[ScoredPoint]):\n",
+    "def rank_list(search_result: list[ScoredPoint]):\n",
     "    return [(point.id, rank + 1) for rank, point in enumerate(search_result)]\n",
     "\n",
     "\n",
@@ -1149,7 +1148,7 @@
    ],
    "source": [
     "def find_point_by_id(\n",
-    "    client: QdrantClient, collection_name: str, rrf_rank_list: List[Tuple[int, float]]\n",
+    "    client: QdrantClient, collection_name: str, rrf_rank_list: list[tuple[int, float]]\n",
     "):\n",
     "    return client.retrieve(\n",
     "        collection_name=collection_name, ids=[item[0] for item in rrf_rank_list]\n",

docs/examples/SPLADE_with_FastEmbed.ipynb

@@ -47,7 +47,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
+   "execution_count": null,
    "metadata": {
     "ExecuteTime": {
      "end_time": "2024-03-30T00:49:20.516644Z",
@@ -56,8 +56,7 @@
    },
    "outputs": [],
    "source": [
-    "from fastembed import SparseTextEmbedding, SparseEmbedding\n",
-    "from typing import List"
+    "from fastembed import SparseTextEmbedding, SparseEmbedding"
    ]
   },
   {
@@ -134,7 +133,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": null,
    "metadata": {
     "ExecuteTime": {
      "end_time": "2024-03-30T00:49:28.624109Z",
@@ -143,7 +142,7 @@
    },
    "outputs": [],
    "source": [
-    "documents: List[str] = [\n",
+    "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",
@@ -157,7 +156,7 @@
     "    \"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",
+    "sparse_embeddings_list: list[SparseEmbedding] = list(\n",
     "    model.embed(documents, batch_size=6)\n",
     ")  # batch_size is optional, notice the generator"
    ]
@@ -235,7 +234,9 @@
    "source": [
     "# Let's print the first 5 features and their weights for better understanding.\n",
     "for i in range(5):\n",
-    "    print(f\"Token at index {sparse_embeddings_list[0].indices[i]} has weight {sparse_embeddings_list[0].values[i]}\")"
+    "    print(\n",
+    "        f\"Token at index {sparse_embeddings_list[0].indices[i]} has weight {sparse_embeddings_list[0].values[i]}\"\n",
+    "    )"
    ]
   },
   {
@@ -261,7 +262,9 @@
     "import json\n",
     "from transformers import AutoTokenizer\n",
     "\n",
-    "tokenizer = AutoTokenizer.from_pretrained(SparseTextEmbedding.list_supported_models()[0][\"sources\"][\"hf\"])"
+    "tokenizer = AutoTokenizer.from_pretrained(\n",
+    "    SparseTextEmbedding.list_supported_models()[0][\"sources\"][\"hf\"]\n",
+    ")"
    ]
   },
   {
@@ -326,7 +329,9 @@
     "        token_weight_dict[token] = weight\n",
     "\n",
     "    # Sort the dictionary by weights\n",
-    "    token_weight_dict = dict(sorted(token_weight_dict.items(), key=lambda item: item[1], reverse=True))\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",

docs/index.md

@@ -26,14 +26,14 @@ pip install fastembed
 ```python
 from fastembed import TextEmbedding
 
-documents: List[str] = [
+documents: list[str] = [
     "passage: Hello, World!",
     "query: Hello, World!",
     "passage: This is an example passage.",
     "fastembed is supported by and maintained by Qdrant."
 ]
 embedding_model = TextEmbedding()
-embeddings: List[np.ndarray] = embedding_model.embed(documents)
+embeddings: list[np.ndarray] = embedding_model.embed(documents)
 ```
 
 ## Usage with Qdrant

docs/qdrant/Retrieval_with_FastEmbed.ipynb

@@ -41,7 +41,6 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "from typing import List\n",
     "import numpy as np\n",
     "from fastembed import TextEmbedding"
    ]
@@ -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",
@@ -87,7 +86,7 @@
     "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",

docs/qdrant/Usage_With_Qdrant.ipynb

@@ -46,7 +46,6 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "from typing import List\n",
     "from qdrant_client import QdrantClient"
    ]
   },
@@ -67,7 +66,7 @@
    "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",
@@ -199,7 +198,9 @@
     }
    ],
    "source": [
-    "search_result = client.query(collection_name=\"demo_collection\", query_text=\"This is a query document\")\n",
+    "search_result = client.query(\n",
+    "    collection_name=\"demo_collection\", query_text=\"This is a query document\"\n",
+    ")\n",
     "print(search_result)"
    ]
   },

experiments/01_ONNX_Port.ipynb

@@ -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",
@@ -91,9 +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",
@@ -133,7 +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",
@@ -171,7 +175,9 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "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",

fastembed/common/model_management.py

@@ -1,29 +1,37 @@
 import os
 import time
+import json
 import shutil
 import tarfile
 from pathlib import Path
-from typing import Any, Dict, List, Optional
+from typing import Any, Optional
 
 import requests
-from huggingface_hub import snapshot_download
-from huggingface_hub.utils import RepositoryNotFoundError
+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
 
 
 class ModelManagement:
+    METADATA_FILE = "files_metadata.json"
+
     @classmethod
-    def list_supported_models(cls) -> List[Dict[str, Any]]:
+    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.
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
         """
         raise NotImplementedError()
 
     @classmethod
-    def _get_model_description(cls, model_name: str) -> Dict[str, Any]:
+    def _get_model_description(cls, model_name: str) -> dict[str, Any]:
         """
         Gets the model description from the model_name.
 
@@ -34,7 +42,7 @@ class ModelManagement:
             ValueError: If the model_name is not supported.
 
         Returns:
-            Dict[str, Any]: The model description.
+            dict[str, Any]: The model description.
         """
         for model in cls.list_supported_models():
             if model_name.lower() == model["model"].lower():
@@ -74,7 +82,7 @@ class ModelManagement:
         if total_size_in_bytes == 0:
             print(f"Warning: Content-length header is missing or zero in the response from {url}.")
 
-        show_progress = total_size_in_bytes and show_progress
+        show_progress = bool(total_size_in_bytes and show_progress)
 
         with tqdm(
             total=total_size_in_bytes,
@@ -93,20 +101,74 @@ class ModelManagement:
     def download_files_from_huggingface(
         cls,
         hf_source_repo: str,
-        cache_dir: Optional[str] = None,
-        extra_patterns: Optional[List[str]] = None,
-        **kwargs,
+        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 (Optional[List[str]]): extra patterns to allow in the snapshot download, typically
+            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.
+            specific_model_path (Optional[str], optional): The path to the model dir already pooled from external source
         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, int]]:
+            meta = {}
+            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, int]]) -> 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",
@@ -114,18 +176,86 @@ class ModelManagement:
             "special_tokens_map.json",
             "preprocessor_config.json",
         ]
-        if extra_patterns is not None:
-            allow_patterns.extend(extra_patterns)
 
-        return snapshot_download(
+        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=kwargs.get("local_files_only", False),
+            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):
+    def decompress_to_cache(cls, targz_path: str, cache_dir: str) -> str:
         """
         Decompresses a .tar.gz file to a cache directory.
 
@@ -162,9 +292,10 @@ class ModelManagement:
         return cache_dir
 
     @classmethod
-    def retrieve_model_gcs(cls, model_name: str, source_url: str, cache_dir: str) -> Path:
+    def retrieve_model_gcs(
+        cls, model_name: str, source_url: str, cache_dir: str, local_files_only: bool = False
+    ) -> Path:
         fast_model_name = f"fast-{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
@@ -183,27 +314,37 @@ class ModelManagement:
         if model_tar_gz.exists():
             model_tar_gz.unlink()
 
-        cls.download_file_from_gcs(
-            source_url,
-            output_path=str(model_tar_gz),
-        )
+        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}"
+            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)
+            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: Dict[str, Any], cache_dir: Path, retries=3, **kwargs) -> Path:
+    def download_model(
+        cls, model: dict[str, Any], cache_dir: Path, retries: int = 3, **kwargs: Any
+    ) -> Path:
         """
         Downloads a model from HuggingFace Hub or Google Cloud Storage.
 
         Args:
-            model (Dict[str, Any]): The model description.
+            model (dict[str, Any]): The model description.
                 Example:
                 ```
                 {
@@ -223,7 +364,11 @@ class ModelManagement:
         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.get("sources", {}).get("hf")
         url_source = model.get("sources", {}).get("url")
 
@@ -241,24 +386,36 @@ class ModelManagement:
                             hf_source,
                             cache_dir=str(cache_dir),
                             extra_patterns=extra_patterns,
-                            local_files_only=kwargs.get("local_files_only", False),
+                            **kwargs,
                         )
                     )
                 except (EnvironmentError, RepositoryNotFoundError, ValueError) as e:
-                    logger.error(
-                        f"Could not download model from HuggingFace: {e} "
-                        "Falling back to other sources."
-                    )
-            if url_source:
+                    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"], url_source, str(cache_dir))
+                    return cls.retrieve_model_gcs(
+                        model["model"],
+                        url_source,
+                        str(cache_dir),
+                        local_files_only=local_files_only,
+                    )
                 except Exception:
-                    logger.error(f"Could not download model from url: {url_source}")
+                    if not local_files_only:
+                        logger.error(f"Could not download model from url: {url_source}")
 
-            logger.error(
-                f"Could not download model from either source, sleeping for {sleep} seconds, {retries} retries left."
-            )
+            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 download model {model['model']} from any source.")
+        raise ValueError(f"Could not load model {model['model']} from any source.")

fastembed/common/onnx_model.py

@@ -1,17 +1,7 @@
 import warnings
 from dataclasses import dataclass
 from pathlib import Path
-from typing import (
-    Any,
-    Dict,
-    Generic,
-    Iterable,
-    Optional,
-    Sequence,
-    Tuple,
-    Type,
-    TypeVar,
-)
+from typing import Any, Generic, Iterable, Optional, Sequence, Type, TypeVar
 
 import numpy as np
 import onnxruntime as ort
@@ -43,26 +33,42 @@ class OnnxModel(Generic[T]):
         self.tokenizer = None
 
     def _preprocess_onnx_input(
-        self, onnx_input: Dict[str, np.ndarray], **kwargs
-    ) -> Dict[str, np.ndarray]:
+        self, onnx_input: dict[str, np.ndarray], **kwargs: Any
+    ) -> dict[str, np.ndarray]:
         """
         Preprocess the onnx input.
         """
         return onnx_input
 
-    def load_onnx_model(
+    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
 
-        onnx_providers = (
-            ["CPUExecutionProvider"] if providers is None else list(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 = []
         for provider in onnx_providers:
@@ -94,7 +100,10 @@ class OnnxModel(Generic[T]):
                     RuntimeWarning,
                 )
 
-    def onnx_embed(self, *args, **kwargs) -> OnnxOutputContext:
+    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")
 
 
@@ -103,7 +112,7 @@ class EmbeddingWorker(Worker):
         self,
         model_name: str,
         cache_dir: str,
-        **kwargs,
+        **kwargs: Any,
     ) -> OnnxModel:
         raise NotImplementedError()
 
@@ -111,7 +120,7 @@ class EmbeddingWorker(Worker):
         self,
         model_name: str,
         cache_dir: str,
-        **kwargs,
+        **kwargs: Any,
     ):
         self.model = self.init_embedding(model_name, cache_dir, **kwargs)
 
@@ -119,5 +128,5 @@ class EmbeddingWorker(Worker):
     def start(cls, model_name: str, cache_dir: str, **kwargs: Any) -> "EmbeddingWorker":
         return cls(model_name=model_name, cache_dir=cache_dir, **kwargs)
 
-    def process(self, items: Iterable[Tuple[int, Any]]) -> Iterable[Tuple[int, Any]]:
+    def process(self, items: Iterable[tuple[int, Any]]) -> Iterable[tuple[int, Any]]:
         raise NotImplementedError("Subclasses must implement this method")

fastembed/common/pooling.py

@@ -0,0 +1,38 @@
+import numpy as np
+from numpy import ufunc
+
+
+class LateInteractionPooler(object):
+    def __init__(self, agg_type="row", agg="mean"):
+        self.agg = agg
+        self.type = agg_type
+
+    def _pick_operation(self) -> ufunc:
+        if self.agg == "mean":
+            return np.mean
+        elif self.agg == "max":
+            return np.max
+        elif self.agg == "min":
+            return np.min
+        else:
+            raise NotImplementedError(
+                f"LateInteractionPooler only supports agg=mean,min,max, provided {self.agg}"
+            )
+
+    def pool(self, embeddings_batch) -> np.array:
+        if isinstance(embeddings_batch, np.ndarray) and len(embeddings_batch.shape) == 2:
+            embeddings_batch = [embeddings_batch]
+
+        if self.type == "row":
+            pooled_embedding = self.pool_row(embeddings_batch)
+        elif self.type == "col":
+            pooled_embedding = self.pool_col(embeddings_batch)
+        else:
+            raise ValueError("type must be 'row' or 'col'")
+        return pooled_embedding
+
+    def pool_row(self, embeddings_batch) -> np.array:
+        return self._pick_operation()(embeddings_batch, axis=-1)
+
+    def pool_col(self, embeddings_batch) -> np.array:
+        return self._pick_operation()(embeddings_batch, axis=-2)

fastembed/common/preprocessor_utils.py

@@ -1,6 +1,6 @@
 import json
 from pathlib import Path
-from typing import Tuple
+
 from tokenizers import AddedToken, Tokenizer
 
 from fastembed.image.transform.operators import Compose
@@ -17,7 +17,7 @@ def load_special_tokens(model_dir: Path) -> dict:
     return tokens_map
 
 
-def load_tokenizer(model_dir: Path) -> Tuple[Tokenizer, dict]:
+def load_tokenizer(model_dir: Path) -> tuple[Tokenizer, dict]:
     config_path = model_dir / "config.json"
     if not config_path.exists():
         raise ValueError(f"Could not find config.json in {model_dir}")

fastembed/common/types.py

@@ -1,7 +1,7 @@
-import os
+from pathlib import Path
 import sys
 from PIL import Image
-from typing import Any, Dict, Iterable, Tuple, Union
+from typing import Any, Iterable, Union
 
 if sys.version_info >= (3, 10):
     from typing import TypeAlias
@@ -9,8 +9,8 @@ else:
     from typing_extensions import TypeAlias
 
 
-PathInput: TypeAlias = Union[str, os.PathLike]
+PathInput: TypeAlias = Union[str, Path]
 PilInput: TypeAlias = Union[Image.Image, Iterable[Image.Image]]
 ImageInput: TypeAlias = Union[PathInput, Iterable[PathInput], PilInput]
 
-OnnxProvider: TypeAlias = Union[str, Tuple[str, Dict[Any, Any]]]
+OnnxProvider: TypeAlias = Union[str, tuple[str, dict[Any, Any]]]

fastembed/common/utils.py

@@ -1,13 +1,18 @@
 import os
+import sys
+import re
 import tempfile
-from itertools import islice
+import unicodedata
 from pathlib import Path
-from typing import Generator, Iterable, Optional, Union
+from itertools import islice
+from typing import Iterable, Optional, TypeVar
 
 import numpy as np
 
+T = TypeVar("T")
 
-def normalize(input_array, p=2, dim=1, eps=1e-12) -> np.ndarray:
+
+def normalize(input_array: np.ndarray, p: int = 2, dim: int = 1, eps: float = 1e-12) -> np.ndarray:
     # 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
@@ -15,7 +20,7 @@ def normalize(input_array, p=2, dim=1, eps=1e-12) -> np.ndarray:
     return normalized_array
 
 
-def iter_batch(iterable: Union[Iterable, Generator], size: int) -> Iterable:
+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]]
@@ -37,7 +42,16 @@ def define_cache_dir(cache_dir: Optional[str] = None) -> Path:
         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)

fastembed/embedding.py

@@ -1,4 +1,4 @@
-from typing import Optional
+from typing import Optional, Any
 
 from loguru import logger
 
@@ -19,6 +19,6 @@ class JinaEmbedding(TextEmbedding):
         model_name: str = "jinaai/jina-embeddings-v2-base-en",
         cache_dir: Optional[str] = None,
         threads: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ):
         super().__init__(model_name, cache_dir, threads, **kwargs)

fastembed/image/image_embedding.py

@@ -1,4 +1,4 @@
-from typing import Any, Dict, Iterable, List, Optional, Sequence, Type
+from typing import Any, Iterable, Optional, Sequence, Type
 
 import numpy as np
 
@@ -8,15 +8,15 @@ from fastembed.image.onnx_embedding import OnnxImageEmbedding
 
 
 class ImageEmbedding(ImageEmbeddingBase):
-    EMBEDDINGS_REGISTRY: List[Type[ImageEmbeddingBase]] = [OnnxImageEmbedding]
+    EMBEDDINGS_REGISTRY: list[Type[ImageEmbeddingBase]] = [OnnxImageEmbedding]
 
     @classmethod
-    def list_supported_models(cls) -> List[Dict[str, Any]]:
+    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.
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
 
             Example:
                 ```
@@ -25,6 +25,7 @@ class ImageEmbedding(ImageEmbeddingBase):
                         "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",
@@ -45,21 +46,23 @@ class ImageEmbedding(ImageEmbeddingBase):
         cache_dir: Optional[str] = None,
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
-        **kwargs,
+        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
-            ):
+            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
@@ -74,7 +77,7 @@ class ImageEmbedding(ImageEmbeddingBase):
         images: ImageInput,
         batch_size: int = 16,
         parallel: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ) -> Iterable[np.ndarray]:
         """
         Encode a list of documents into list of embeddings.

fastembed/image/image_embedding_base.py

@@ -1,4 +1,4 @@
-from typing import Iterable, Optional
+from typing import Iterable, Optional, Any
 
 import numpy as np
 
@@ -12,7 +12,7 @@ class ImageEmbeddingBase(ModelManagement):
         model_name: str,
         cache_dir: Optional[str] = None,
         threads: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ):
         self.model_name = model_name
         self.cache_dir = cache_dir
@@ -24,13 +24,13 @@ class ImageEmbeddingBase(ModelManagement):
         images: ImageInput,
         batch_size: int = 16,
         parallel: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ) -> Iterable[np.ndarray]:
         """
         Embeds a list of images into a list of embeddings.
 
         Args:
-            images - The list of image paths to preprocess and embed.
+            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.

fastembed/image/onnx_embedding.py

@@ -1,7 +1,6 @@
-from typing import Any, Dict, Iterable, List, Optional, Sequence, Type
+from typing import Any, Iterable, Optional, Sequence, Type
 
 import numpy as np
-
 from fastembed.common import ImageInput, OnnxProvider
 from fastembed.common.onnx_model import OnnxOutputContext
 from fastembed.common.utils import define_cache_dir, normalize
@@ -13,6 +12,7 @@ supported_onnx_models = [
         "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": {
             "hf": "Qdrant/clip-ViT-B-32-vision",
@@ -23,6 +23,7 @@ supported_onnx_models = [
         "model": "Qdrant/resnet50-onnx",
         "dim": 2048,
         "description": "Image embeddings, Unimodal (image), 2016 year",
+        "license": "apache-2.0",
         "size_in_GB": 0.1,
         "sources": {
             "hf": "Qdrant/resnet50-onnx",
@@ -33,6 +34,7 @@ supported_onnx_models = [
         "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": {
             "hf": "Qdrant/Unicom-ViT-B-16",
@@ -43,12 +45,24 @@ supported_onnx_models = [
         "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": {
             "hf": "Qdrant/Unicom-ViT-B-32",
         },
         "model_file": "model.onnx",
     },
+    {
+        "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": {
+            "hf": "jinaai/jina-clip-v1",
+        },
+        "model_file": "onnx/vision_model.onnx",
+    },
 ]
 
 
@@ -59,7 +73,12 @@ class OnnxImageEmbedding(ImageEmbeddingBase, OnnxImageModel[np.ndarray]):
         cache_dir: Optional[str] = None,
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
-        **kwargs,
+        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:
@@ -68,33 +87,69 @@ class OnnxImageEmbedding(ImageEmbeddingBase, OnnxImageModel[np.ndarray]):
                                        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
 
-        model_description = self._get_model_description(model_name)
+        # 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
+        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]
+        else:
+            self.device_id = None
+
+        self.model_description = self._get_model_description(model_name)
         self.cache_dir = define_cache_dir(cache_dir)
-        model_dir = self.download_model(
-            model_description, self.cache_dir, local_files_only=self._local_files_only
+        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.load_onnx_model(
-            model_dir=model_dir,
-            model_file=model_description["model_file"],
-            threads=threads,
-            providers=providers,
+        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[Dict[str, Any]]:
+    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.
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
         """
         return supported_onnx_models
 
@@ -103,7 +158,7 @@ class OnnxImageEmbedding(ImageEmbeddingBase, OnnxImageModel[np.ndarray]):
         images: ImageInput,
         batch_size: int = 16,
         parallel: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ) -> Iterable[np.ndarray]:
         """
         Encode a list of images into list of embeddings.
@@ -120,12 +175,16 @@ class OnnxImageEmbedding(ImageEmbeddingBase, OnnxImageModel[np.ndarray]):
         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,
         )
 
@@ -134,8 +193,8 @@ class OnnxImageEmbedding(ImageEmbeddingBase, OnnxImageModel[np.ndarray]):
         return OnnxImageEmbeddingWorker
 
     def _preprocess_onnx_input(
-        self, onnx_input: Dict[str, np.ndarray], **kwargs
-    ) -> Dict[str, np.ndarray]:
+        self, onnx_input: dict[str, np.ndarray], **kwargs: Any
+    ) -> dict[str, np.ndarray]:
         """
         Preprocess the onnx input.
         """
@@ -147,5 +206,10 @@ class OnnxImageEmbedding(ImageEmbeddingBase, OnnxImageModel[np.ndarray]):
 
 
 class OnnxImageEmbeddingWorker(ImageEmbeddingWorker):
-    def init_embedding(self, model_name: str, cache_dir: str, **kwargs) -> OnnxImageEmbedding:
-        return OnnxImageEmbedding(model_name=model_name, cache_dir=cache_dir, threads=1, **kwargs)
+    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,
+        )

fastembed/image/onnx_image_model.py

@@ -2,7 +2,7 @@ import contextlib
 import os
 from multiprocessing import get_all_start_methods
 from pathlib import Path
-from typing import Any, Dict, Iterable, List, Optional, Sequence, Tuple, Type
+from typing import Any, Iterable, Optional, Sequence, Type
 
 import numpy as np
 from PIL import Image
@@ -24,37 +24,44 @@ class OnnxImageModel(OnnxModel[T]):
     def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[T]:
         raise NotImplementedError("Subclasses must implement this method")
 
-    def __init__(self) -> None:
+    def __init__(self):
         super().__init__()
         self.processor = None
 
     def _preprocess_onnx_input(
-        self, onnx_input: Dict[str, np.ndarray], **kwargs
-    ) -> Dict[str, np.ndarray]:
+        self, onnx_input: dict[str, np.ndarray], **kwargs: Any
+    ) -> dict[str, np.ndarray]:
         """
         Preprocess the onnx input.
         """
         return onnx_input
 
-    def load_onnx_model(
+    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(
+        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 _build_onnx_input(self, encoded: np.ndarray) -> Dict[str, np.ndarray]:
+    def load_onnx_model(self) -> None:
+        raise NotImplementedError("Subclasses must implement this method")
+
+    def _build_onnx_input(self, encoded: np.ndarray) -> dict[str, np.ndarray]:
         return {node.name: encoded for node in self.model.get_inputs()}
 
-    def onnx_embed(self, images: List[ImageInput], **kwargs) -> OnnxOutputContext:
+    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
@@ -74,40 +81,51 @@ class OnnxImageModel(OnnxModel[T]):
         images: ImageInput,
         batch_size: int = 256,
         parallel: Optional[int] = None,
-        **kwargs,
+        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)
-            or isinstance(images, Path)
-            or (isinstance(images, Image.Image))
-        ):
+        if isinstance(images, (str, Path, Image.Image)):
             images = [images]
             is_small = True
 
-        if isinstance(images, list):
-            if len(images) < batch_size:
-                is_small = True
-
-        if parallel == 0:
-            parallel = os.cpu_count()
+        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, **kwargs}
+            params = {
+                "model_name": model_name,
+                "cache_dir": cache_dir,
+                "providers": providers,
+                **kwargs,
+            }
+
             pool = ParallelWorkerPool(
-                parallel, self._get_worker_class(), start_method=start_method
+                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)
 
 
 class ImageEmbeddingWorker(EmbeddingWorker):
-    def process(self, items: Iterable[Tuple[int, Any]]) -> Iterable[Tuple[int, Any]]:
+    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

fastembed/image/transform/functional.py

@@ -1,4 +1,4 @@
-from typing import Sized, Tuple, Union
+from typing import Sized, Union
 
 import numpy as np
 from PIL import Image
@@ -14,7 +14,7 @@ def convert_to_rgb(image: Image.Image) -> Image.Image:
 
 def center_crop(
     image: Union[Image.Image, np.ndarray],
-    size: Tuple[int, int],
+    size: tuple[int, int],
 ) -> np.ndarray:
     if isinstance(image, np.ndarray):
         _, orig_height, orig_width = image.shape
@@ -62,8 +62,8 @@ def center_crop(
 
 def normalize(
     image: np.ndarray,
-    mean=Union[float, np.ndarray],
-    std=Union[float, np.ndarray],
+    mean: Union[float, np.ndarray],
+    std: Union[float, np.ndarray],
 ) -> np.ndarray:
     if not isinstance(image, np.ndarray):
         raise ValueError("image must be a numpy array")
@@ -96,10 +96,10 @@ def normalize(
 
 
 def resize(
-    image: Image,
-    size: Union[int, Tuple[int, int]],
-    resample: Image.Resampling = Image.Resampling.BILINEAR,
-) -> Image:
+    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)
 
@@ -114,11 +114,37 @@ def resize(
     return image.resize(new_size, resample)
 
 
-def rescale(image: np.ndarray, scale: float, dtype=np.float32) -> np.ndarray:
+def rescale(image: np.ndarray, scale: float, dtype: type = np.float32) -> np.ndarray:
     return (image * scale).astype(dtype)
 
 
-def pil2ndarray(image: Union[Image.Image, np.ndarray]):
+def pil2ndarray(image: Union[Image.Image, np.ndarray]) -> np.ndarray:
     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

fastembed/image/transform/operators.py

@@ -1,4 +1,4 @@
-from typing import Any, Dict, List, Tuple, Union
+from typing import Any, Union, Optional
 
 import numpy as np
 from PIL import Image
@@ -10,93 +10,111 @@ from fastembed.image.transform.functional import (
     pil2ndarray,
     rescale,
     resize,
+    pad2square,
 )
 
 
 class Transform:
-    def __call__(self, images: List) -> Union[List[Image.Image], List[np.ndarray]]:
+    def __call__(self, images: list) -> Union[list[Image.Image], list[np.ndarray]]:
         raise NotImplementedError("Subclasses must implement this method")
 
 
 class ConvertToRGB(Transform):
-    def __call__(self, images: List[Image.Image]) -> List[Image.Image]:
+    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]):
+    def __init__(self, size: tuple[int, int]):
         self.size = size
 
-    def __call__(self, images: List[Image.Image]) -> List[np.ndarray]:
+    def __call__(self, images: list[Image.Image]) -> list[np.ndarray]:
         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]]):
+    def __init__(self, mean: Union[float, list[float]], std: Union[float, list[float]]):
         self.mean = mean
         self.std = std
 
-    def __call__(self, images: List[np.ndarray]) -> List[np.ndarray]:
+    def __call__(self, images: list[np.ndarray]) -> list[np.ndarray]:
         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]],
+        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
-        ]
+    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[np.ndarray]) -> List[np.ndarray]:
+    def __call__(self, images: list[np.ndarray]) -> list[np.ndarray]:
         return [rescale(image, scale=self.scale) for image in images]
 
 
 class PILtoNDarray(Transform):
-    def __call__(
-        self, images: List[Union[Image.Image, np.ndarray]]
-    ) -> List[np.ndarray]:
+    def __call__(self, images: list[Union[Image.Image, np.ndarray]]) -> list[np.ndarray]:
         return [pil2ndarray(image) for image in images]
 
 
+class PadtoSquare(Transform):
+    def __init__(
+        self,
+        size: int,
+        fill_color: Optional[Union[str, int, tuple[int, ...]]] = None,
+    ):
+        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]):
+    def __init__(self, transforms: list[Transform]):
         self.transforms = transforms
 
     def __call__(
-        self, images: Union[List[Image.Image], List[np.ndarray]]
-    ) -> Union[List[np.ndarray], List[Image.Image]]:
+        self, images: Union[list[Image.Image], list[np.ndarray]]
+    ) -> Union[list[np.ndarray], list[Image.Image]]:
         for transform in self.transforms:
             images = transform(images)
         return images
 
     @classmethod
-    def from_config(cls, config: Dict[str, Any]) -> "Compose":
+    def from_config(cls, config: dict[str, Any]) -> "Compose":
         """Creates processor from a config dict.
         Args:
-            config (Dict[str, Any]): Configuration dictionary.
+            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"}
@@ -107,6 +125,7 @@ class Compose:
         transforms = []
         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)
@@ -114,11 +133,11 @@ class Compose:
         return cls(transforms=transforms)
 
     @staticmethod
-    def _get_convert_to_rgb(transforms: List[Transform], config: Dict[str, Any]):
+    def _get_convert_to_rgb(transforms: list[Transform], config: dict[str, Any]) -> None:
         transforms.append(ConvertToRGB())
 
-    @staticmethod
-    def _get_resize(transforms: List[Transform], config: Dict[str, Any]):
+    @classmethod
+    def _get_resize(cls, transforms: list[Transform], config: dict[str, Any]) -> None:
         mode = config.get("image_processor_type", "CLIPImageProcessor")
         if mode == "CLIPImageProcessor":
             if config.get("do_resize", False):
@@ -161,9 +180,27 @@ class Compose:
                         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]):
+    def _get_center_crop(transforms: list[Transform], config: dict[str, Any]) -> None:
         mode = config.get("image_processor_type", "CLIPImageProcessor")
         if mode == "CLIPImageProcessor":
             if config.get("do_center_crop", False):
@@ -177,22 +214,55 @@ class Compose:
                 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]):
+    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]):
+    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]):
+    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(
-                Normalize(mean=config["image_mean"], std=config["image_std"])
+                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}")

fastembed/late_interaction/colbert.py

@@ -1,5 +1,5 @@
 import string
-from typing import Any, Dict, Iterable, List, Optional, Sequence, Type, Union
+from typing import Any, Iterable, Optional, Sequence, Type, Union
 
 import numpy as np
 from tokenizers import Encoding
@@ -12,11 +12,13 @@ from fastembed.late_interaction.late_interaction_embedding_base import (
 )
 from fastembed.text.onnx_text_model import OnnxTextModel, TextEmbeddingWorker
 
+
 supported_colbert_models = [
     {
         "model": "colbert-ir/colbertv2.0",
         "dim": 128,
         "description": "Late interaction model",
+        "license": "mit",
         "size_in_GB": 0.44,
         "sources": {
             "hf": "colbert-ir/colbertv2.0",
@@ -27,6 +29,7 @@ supported_colbert_models = [
         "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": {
             "hf": "answerdotai/answerai-colbert-small-v1",
@@ -39,7 +42,7 @@ supported_colbert_models = [
 class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[np.ndarray]):
     QUERY_MARKER_TOKEN_ID = 1
     DOCUMENT_MARKER_TOKEN_ID = 2
-    MIN_QUERY_LENGTH = 32
+    MIN_QUERY_LENGTH = 31  # it's 32, we add one additional special token in the beginning
     MASK_TOKEN = "[MASK]"
 
     def _post_process_onnx_output(
@@ -48,6 +51,11 @@ class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[np.ndarray]):
         if not is_doc:
             return output.model_output.astype(np.float32)
 
+        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):
                 if token_id in self.skip_list or token_id == self.pad_token_id:
@@ -60,25 +68,22 @@ class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[np.ndarray]):
         return output.model_output.astype(np.float32)
 
     def _preprocess_onnx_input(
-        self, onnx_input: Dict[str, np.ndarray], is_doc: bool = True
-    ) -> Dict[str, np.ndarray]:
-        if is_doc:
-            onnx_input["input_ids"][:, 1] = self.DOCUMENT_MARKER_TOKEN_ID
-        else:
-            onnx_input["input_ids"][:, 1] = self.QUERY_MARKER_TOKEN_ID
+        self, onnx_input: dict[str, np.ndarray], is_doc: bool = True, **kwargs: Any
+    ) -> dict[str, np.ndarray]:
+        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"], 1, marker_token, axis=1)
+        onnx_input["attention_mask"] = np.insert(onnx_input["attention_mask"], 1, 1, axis=1)
         return onnx_input
 
-    def tokenize(self, documents: List[str], is_doc: bool = True) -> List[Encoding]:
+    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]:
-        # ". " is added to a query to be replaced with a special query token
-        query = [f". {query}"]
-        encoded = self.tokenizer.encode_batch(query)
+    def _tokenize_query(self, query: str) -> list[Encoding]:
+        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
@@ -89,25 +94,23 @@ class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[np.ndarray]):
                 pad_id=self.mask_token_id,
                 length=self.MIN_QUERY_LENGTH,
             )
-            encoded = self.tokenizer.encode_batch(query)
+            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]:
-        # ". " is added to a document to be replaced with a special document token
-        documents = [". " + doc for doc in documents]
+    def _tokenize_documents(self, documents: list[str]) -> list[Encoding]:
         encoded = self.tokenizer.encode_batch(documents)
         return encoded
 
     @classmethod
-    def list_supported_models(cls) -> List[Dict[str, Any]]:
+    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.
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
         """
         return supported_colbert_models
 
@@ -117,7 +120,12 @@ class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[np.ndarray]):
         cache_dir: Optional[str] = None,
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
-        **kwargs,
+        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:
@@ -126,40 +134,78 @@ class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[np.ndarray]):
                                        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
 
-        model_description = self._get_model_description(model_name)
+        # 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
+        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]
+        else:
+            self.device_id = None
+
+        self.model_description = self._get_model_description(model_name)
         self.cache_dir = define_cache_dir(cache_dir)
 
-        model_dir = self.download_model(
-            model_description, self.cache_dir, local_files_only=self._local_files_only
+        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
+        self.skip_list = set()
 
-        self.load_onnx_model(
-            model_dir=model_dir,
-            model_file=model_description["model_file"],
-            threads=threads,
-            providers=providers,
+        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,
         )
-        self.mask_token_id = self.special_token_to_id["[MASK]"]
+        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,
+        **kwargs: Any,
     ) -> Iterable[np.ndarray]:
         """
         Encode a list of documents into list of embeddings.
@@ -182,13 +228,19 @@ class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[np.ndarray]):
             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, List[str]], **kwargs) -> np.ndarray:
+    def query_embed(self, query: Union[str, Iterable[str]], **kwargs: Any) -> Iterable[np.ndarray]:
         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
@@ -200,5 +252,10 @@ class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[np.ndarray]):
 
 
 class ColbertEmbeddingWorker(TextEmbeddingWorker):
-    def init_embedding(self, model_name: str, cache_dir: str, **kwargs) -> Colbert:
-        return Colbert(model_name=model_name, cache_dir=cache_dir, threads=1, **kwargs)
+    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,
+        )

fastembed/late_interaction/jina_colbert.py

@@ -0,0 +1,62 @@
+from typing import Any, Type
+
+import numpy as np
+
+from fastembed.late_interaction.colbert import Colbert
+from fastembed.text.onnx_text_model import TextEmbeddingWorker
+
+
+supported_jina_colbert_models = [
+    {
+        "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": {
+            "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[TextEmbeddingWorker]:
+        return JinaColbertEmbeddingWorker
+
+    @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 supported_jina_colbert_models
+
+    def _preprocess_onnx_input(
+        self, onnx_input: dict[str, np.ndarray], is_doc: bool = True, **kwargs: Any
+    ) -> dict[str, np.ndarray]:
+        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(TextEmbeddingWorker):
+    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,
+        )

fastembed/late_interaction/late_interaction_embedding_base.py

@@ -1,4 +1,4 @@
-from typing import Iterable, Optional, Union
+from typing import Iterable, Optional, Union, Any
 
 import numpy as np
 
@@ -11,7 +11,7 @@ class LateInteractionTextEmbeddingBase(ModelManagement):
         model_name: str,
         cache_dir: Optional[str] = None,
         threads: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ):
         self.model_name = model_name
         self.cache_dir = cache_dir
@@ -23,11 +23,11 @@ class LateInteractionTextEmbeddingBase(ModelManagement):
         documents: Union[str, Iterable[str]],
         batch_size: int = 256,
         parallel: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ) -> Iterable[np.ndarray]:
         raise NotImplementedError()
 
-    def passage_embed(self, texts: Iterable[str], **kwargs) -> Iterable[np.ndarray]:
+    def passage_embed(self, texts: Iterable[str], **kwargs: Any) -> Iterable[np.ndarray]:
         """
         Embeds a list of text passages into a list of embeddings.
 
@@ -42,9 +42,7 @@ class LateInteractionTextEmbeddingBase(ModelManagement):
         # 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
-    ) -> Iterable[np.ndarray]:
+    def query_embed(self, query: Union[str, Iterable[str]], **kwargs: Any) -> Iterable[np.ndarray]:
         """
         Embeds queries
 

fastembed/late_interaction/late_interaction_text_embedding.py

@@ -1,39 +1,40 @@
-from typing import Any, Dict, Iterable, List, Optional, Sequence, Type, Union
+from typing import Any, Iterable, Optional, Sequence, Type, Union
 
 import numpy as np
 
 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,
-    ]
+    EMBEDDINGS_REGISTRY: list[Type[LateInteractionTextEmbeddingBase]] = [Colbert, JinaColbert]
 
     @classmethod
-    def list_supported_models(cls) -> List[Dict[str, Any]]:
+    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.
+            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",
-                        "size_in_GB": 0.532,
+                        "model": "colbert-ir/colbertv2.0",
+                        "dim": 128,
+                        "description": "Late interaction model",
+                        "license": "mit",
+                        "size_in_GB": 0.44,
                         "sources": {
-                            "hf": "qdrant/SPLADE_PP_en_v1",
+                            "hf": "colbert-ir/colbertv2.0",
                         },
-                    }
+                        "model_file": "model.onnx",
+                    },
                 ]
                 ```
         """
@@ -48,24 +49,30 @@ class LateInteractionTextEmbedding(LateInteractionTextEmbeddingBase):
         cache_dir: Optional[str] = None,
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
-        **kwargs,
+        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
-            ):
+            if any(model_name.lower() == model["model"].lower() for model in supported_models):
                 self.model = EMBEDDING_MODEL_TYPE(
-                    model_name, cache_dir, threads, providers=providers, **kwargs
+                    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()`"
+            f"Model {model_name} is not supported in LateInteractionTextEmbedding."
+            "Please check the supported models using `LateInteractionTextEmbedding.list_supported_models()`"
         )
 
     def embed(
@@ -73,7 +80,7 @@ class LateInteractionTextEmbedding(LateInteractionTextEmbeddingBase):
         documents: Union[str, Iterable[str]],
         batch_size: int = 256,
         parallel: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ) -> Iterable[np.ndarray]:
         """
         Encode a list of documents into list of embeddings.
@@ -92,9 +99,7 @@ class LateInteractionTextEmbedding(LateInteractionTextEmbeddingBase):
         """
         yield from self.model.embed(documents, batch_size, parallel, **kwargs)
 
-    def query_embed(
-        self, query: Union[str, Iterable[str]], **kwargs
-    ) -> Iterable[np.ndarray]:
+    def query_embed(self, query: Union[str, Iterable[str]], **kwargs: Any) -> Iterable[np.ndarray]:
         """
         Embeds queries
 

fastembed/parallel_processor.py

@@ -1,13 +1,15 @@
 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, Dict, Iterable, List, Optional, Tuple, Type
+from typing import Any, Iterable, Optional, Type
+
 
 # Single item should be processed in less than:
 processing_timeout = 10 * 60  # seconds
@@ -23,10 +25,10 @@ class QueueSignals(str, Enum):
 
 class Worker:
     @classmethod
-    def start(cls, **kwargs: Any) -> "Worker":
+    def start(cls, *args: Any, **kwargs: Any) -> "Worker":
         raise NotImplementedError()
 
-    def process(self, items: Iterable[Tuple[int, Any]]) -> Iterable[Tuple[int, Any]]:
+    def process(self, items: Iterable[tuple[int, Any]]) -> Iterable[tuple[int, Any]]:
         raise NotImplementedError()
 
 
@@ -36,7 +38,7 @@ def _worker(
     output_queue: Queue,
     num_active_workers: BaseValue,
     worker_id: int,
-    kwargs: Optional[Dict[str, Any]] = None,
+    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.
@@ -47,7 +49,9 @@ def _worker(
     if kwargs is None:
         kwargs = {}
 
-    logging.info(f"Reader worker: {worker_id} PID: {os.getpid()}")
+    logging.info(
+        f"Reader worker: {worker_id} PID: {os.getpid()} Device: {kwargs.get('device_id', 'CPU')}"
+    )
     try:
         worker = worker_class.start(**kwargs)
 
@@ -73,7 +77,9 @@ def _worker(
         # 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():
@@ -84,16 +90,23 @@ def _worker(
 
 class ParallelWorkerPool:
     def __init__(
-        self, num_workers: int, worker: Type[Worker], start_method: Optional[str] = None
+        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.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:
@@ -105,6 +118,12 @@ class ParallelWorkerPool:
         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,
@@ -114,15 +133,13 @@ class ParallelWorkerPool:
                     self.output_queue,
                     self.num_active_workers,
                     worker_id,
-                    kwargs.copy(),
+                    worker_kwargs,
                 ),
             )
             process.start()
             self.processes.append(process)
 
-    def ordered_map(
-        self, stream: Iterable[Any], *args: Any, **kwargs: Any
-    ) -> Iterable[Any]:
+    def ordered_map(self, stream: Iterable[Any], *args: Any, **kwargs: Any) -> Iterable[Any]:
         buffer = defaultdict(Any)
         next_expected = 0
 
@@ -134,7 +151,7 @@ class ParallelWorkerPool:
 
     def semi_ordered_map(
         self, stream: Iterable[Any], *args: Any, **kwargs: Any
-    ) -> Iterable[Tuple[int, Any]]:
+    ) -> Iterable[tuple[int, Any]]:
         try:
             self.start(**kwargs)
 
@@ -144,6 +161,7 @@ class ParallelWorkerPool:
             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()
@@ -170,6 +188,7 @@ class ParallelWorkerPool:
                 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()
@@ -179,8 +198,27 @@ class ParallelWorkerPool:
         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:
         """
@@ -210,4 +248,5 @@ class ParallelWorkerPool:
         https://eli.thegreenplace.net/2009/06/12/safely-using-destructors-in-python/.
         """
         for process in self.processes:
-            process.terminate()
+            if process.is_alive():
+                process.terminate()

fastembed/rerank/cross_encoder/__init__.py

@@ -0,0 +1,3 @@
+from fastembed.rerank.cross_encoder.text_cross_encoder import TextCrossEncoder
+
+__all__ = ["TextCrossEncoder"]

fastembed/rerank/cross_encoder/onnx_text_cross_encoder.py

@@ -0,0 +1,227 @@
+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
+
+supported_onnx_models = [
+    {
+        "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",
+    },
+    {
+        "model": "Xenova/ms-marco-MiniLM-L-12-v2",
+        "size_in_GB": 0.12,
+        "sources": {
+            "hf": "Xenova/ms-marco-MiniLM-L-12-v2",
+        },
+        "model_file": "onnx/model.onnx",
+        "description": "MiniLM-L-12-v2 model optimized for re-ranking tasks.",
+        "license": "apache-2.0",
+    },
+    {
+        "model": "BAAI/bge-reranker-base",
+        "size_in_GB": 1.04,
+        "sources": {
+            "hf": "BAAI/bge-reranker-base",
+        },
+        "model_file": "onnx/model.onnx",
+        "description": "BGE reranker base model for cross-encoder re-ranking.",
+        "license": "mit",
+    },
+    {
+        "model": "jinaai/jina-reranker-v1-tiny-en",
+        "size_in_GB": 0.13,
+        "sources": {
+            "hf": "jinaai/jina-reranker-v1-tiny-en",
+        },
+        "model_file": "onnx/model.onnx",
+        "description": "Designed for blazing-fast re-ranking with 8K context length and fewer parameters than jina-reranker-v1-turbo-en.",
+        "license": "apache-2.0",
+    },
+    {
+        "model": "jinaai/jina-reranker-v1-turbo-en",
+        "size_in_GB": 0.15,
+        "sources": {
+            "hf": "jinaai/jina-reranker-v1-turbo-en",
+        },
+        "model_file": "onnx/model.onnx",
+        "description": "Designed for blazing-fast re-ranking with 8K context length.",
+        "license": "apache-2.0",
+    },
+    {
+        "model": "jinaai/jina-reranker-v2-base-multilingual",
+        "size_in_GB": 1.11,
+        "sources": {
+            "hf": "jinaai/jina-reranker-v2-base-multilingual",
+        },
+        "model_file": "onnx/model.onnx",
+        "description": "A multi-lingual reranker model for cross-encoder re-ranking with 1K context length and sliding window",
+        "license": "cc-by-nc-4.0",
+    },
+]
+
+
+class OnnxTextCrossEncoder(TextCrossEncoderBase, OnnxCrossEncoderModel):
+    @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 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
+        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]
+        else:
+            self.device_id = None
+
+        self.model_description = self._get_model_description(model_name)
+        self.cache_dir = 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,
+        )

fastembed/rerank/cross_encoder/onnx_text_model.py

@@ -0,0 +1,153 @@
+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
+
+from fastembed.common.onnx_model import (
+    EmbeddingWorker,
+    OnnxModel,
+    OnnxOutputContext,
+    OnnxProvider,
+)
+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)
+
+    def tokenize(self, pairs: list[tuple[str, str]], **_: Any) -> list[Encoding]:
+        return self.tokenizer.encode_batch(pairs)
+
+    def _build_onnx_input(
+        self, tokenized_input
+    ) -> dict[str, NDArray[Union[np.float32, np.int64]]]:
+        input_names = {node.name for node in self.model.get_inputs()}
+        inputs = {
+            "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)
+        relevant_output = outputs[0]
+        scores = 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)
+
+    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, np.ndarray], **kwargs: Any
+    ) -> dict[str, np.ndarray]:
+        """
+        Preprocess the onnx input.
+        """
+        return onnx_input
+
+
+class TextRerankerWorker(EmbeddingWorker):
+    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

fastembed/rerank/cross_encoder/text_cross_encoder.py

@@ -0,0 +1,120 @@
+from typing import Any, Iterable, Optional, Sequence, Type
+
+from fastembed.common import OnnxProvider
+from fastembed.rerank.cross_encoder.onnx_text_cross_encoder import OnnxTextCrossEncoder
+from fastembed.rerank.cross_encoder.text_cross_encoder_base import TextCrossEncoderBase
+
+
+class TextCrossEncoder(TextCrossEncoderBase):
+    CROSS_ENCODER_REGISTRY: list[Type[TextCrossEncoderBase]] = [
+        OnnxTextCrossEncoder,
+    ]
+
+    @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": "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",
+                    }
+                ]
+                ```
+        """
+        result = []
+        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
+        )

fastembed/rerank/cross_encoder/text_cross_encoder_base.py

@@ -0,0 +1,58 @@
+from typing import Any, Iterable, Optional
+
+from fastembed.common.model_management import ModelManagement
+
+
+class TextCrossEncoderBase(ModelManagement):
+    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")

fastembed/sparse/bm25.py

@@ -1,21 +1,24 @@
 import os
-import string
 from collections import defaultdict
 from multiprocessing import get_all_start_methods
 from pathlib import Path
-from typing import Any, Dict, Iterable, List, Optional, Tuple, Type, Union
+from typing import Any, Iterable, Optional, Type, Union
 
 import mmh3
 import numpy as np
-from snowballstemmer import stemmer as get_stemmer
-
-from fastembed.common.utils import define_cache_dir, iter_batch
+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 WordTokenizer
+from fastembed.sparse.utils.tokenizer import SimpleTokenizer
 
 supported_languages = [
     "arabic",
@@ -53,6 +56,7 @@ supported_bm25_models = [
     {
         "model": "Qdrant/bm25",
         "description": "BM25 as sparse embeddings meant to be used with Qdrant",
+        "license": "apache-2.0",
         "size_in_GB": 0.01,
         "sources": {
             "hf": "Qdrant/bm25",
@@ -88,6 +92,8 @@ class Bm25(SparseTextEmbeddingBase):
         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.
     """
@@ -100,7 +106,10 @@ class Bm25(SparseTextEmbeddingBase):
         b: float = 0.75,
         avg_len: float = 256.0,
         language: str = "english",
-        **kwargs,
+        token_max_length: int = 40,
+        disable_stemmer: bool = False,
+        specific_model_path: Optional[str] = None,
+        **kwargs: Any,
     ):
         super().__init__(model_name, cache_dir, **kwargs)
 
@@ -116,26 +125,37 @@ class Bm25(SparseTextEmbeddingBase):
         model_description = self._get_model_description(model_name)
         self.cache_dir = define_cache_dir(cache_dir)
 
-        model_dir = self.download_model(
-            model_description, self.cache_dir, local_files_only=self._local_files_only
+        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.punctuation = set(string.punctuation)
-        self.stopwords = set(self._load_stopwords(model_dir, self.language))
-        self.stemmer = get_stemmer(language)
-        self.tokenizer = WordTokenizer
+        self.token_max_length = token_max_length
+        self.punctuation = set(get_all_punctuation())
+        self.disable_stemmer = disable_stemmer
+
+        if disable_stemmer:
+            self.stopwords = 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[Dict[str, Any]]:
+    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.
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
         """
         return supported_bm25_models
 
     @classmethod
-    def _load_stopwords(cls, model_dir: Path, language: str) -> List[str]:
+    def _load_stopwords(cls, model_dir: Path, language: str) -> list[str]:
         stopwords_path = model_dir / f"{language}.txt"
         if not stopwords_path.exists():
             return []
@@ -161,13 +181,13 @@ class Bm25(SparseTextEmbeddingBase):
             if len(documents) < batch_size:
                 is_small = True
 
-        if parallel == 0:
-            parallel = os.cpu_count()
-
         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,
@@ -175,9 +195,14 @@ class Bm25(SparseTextEmbeddingBase):
                 "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(
-                parallel, self._get_worker_class(), start_method=start_method
+                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:
@@ -188,7 +213,7 @@ class Bm25(SparseTextEmbeddingBase):
         documents: Union[str, Iterable[str]],
         batch_size: int = 256,
         parallel: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ) -> Iterable[SparseEmbedding]:
         """
         Encode a list of documents into list of embeddings.
@@ -213,16 +238,21 @@ class Bm25(SparseTextEmbeddingBase):
             parallel=parallel,
         )
 
-    def _stem(self, tokens: List[str]) -> List[str]:
+    def _stem(self, tokens: list[str]) -> list[str]:
         stemmed_tokens = []
         for token in tokens:
+            lower_token = token.lower()
+
             if token in self.punctuation:
                 continue
 
-            if token in self.stopwords:
+            if lower_token in self.stopwords:
                 continue
 
-            stemmed_token = self.stemmer.stemWord(token)
+            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)
@@ -230,17 +260,18 @@ class Bm25(SparseTextEmbeddingBase):
 
     def raw_embed(
         self,
-        documents: List[str],
-    ) -> List[SparseEmbedding]:
+        documents: list[str],
+    ) -> list[SparseEmbedding]:
         embeddings = []
         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]:
+    def _term_frequency(self, tokens: list[str]) -> dict[int, float]:
         """Calculate the term frequency part of the BM25 formula.
 
         (
@@ -250,10 +281,10 @@ class Bm25(SparseTextEmbeddingBase):
         )
 
         Args:
-            tokens (List[str]): The list of tokens in the document.
+            tokens (list[str]): The list of tokens in the document.
 
         Returns:
-            Dict[int, float]: The token_id to term frequency mapping.
+            dict[int, float]: The token_id to term frequency mapping.
         """
         tf_map = {}
         counter = defaultdict(int)
@@ -274,7 +305,9 @@ class Bm25(SparseTextEmbeddingBase):
     def compute_token_id(cls, token: str) -> int:
         return abs(mmh3.hash(token))
 
-    def query_embed(self, query: Union[str, Iterable[str]], **kwargs) -> Iterable[SparseEmbedding]:
+    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.
         """
@@ -282,6 +315,7 @@ class Bm25(SparseTextEmbeddingBase):
             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(
@@ -301,7 +335,7 @@ class Bm25Worker(Worker):
         self,
         model_name: str,
         cache_dir: str,
-        **kwargs,
+        **kwargs: Any,
     ):
         self.model = self.init_embedding(model_name, cache_dir, **kwargs)
 
@@ -309,11 +343,11 @@ class Bm25Worker(Worker):
     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, Any]]:
+    def process(self, items: Iterable[tuple[int, Any]]) -> Iterable[tuple[int, Any]]:
         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) -> Bm25:
+    def init_embedding(model_name: str, cache_dir: str, **kwargs: Any) -> Bm25:
         return Bm25(model_name=model_name, cache_dir=cache_dir, **kwargs)

fastembed/sparse/bm42.py

@@ -1,11 +1,11 @@
 import math
 import string
 from pathlib import Path
-from typing import Any, Dict, Iterable, List, Optional, Sequence, Tuple, Type, Union
+from typing import Any, Iterable, Optional, Sequence, Type, Union
 
 import mmh3
 import numpy as np
-from snowballstemmer import stemmer as get_stemmer
+from py_rust_stemmers import SnowballStemmer
 
 from fastembed.common import OnnxProvider
 from fastembed.common.onnx_model import OnnxOutputContext
@@ -21,6 +21,7 @@ supported_bm42_models = [
         "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": {
             "hf": "Qdrant/all_miniLM_L6_v2_with_attentions",
@@ -61,7 +62,12 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
         alpha: float = 0.5,
-        **kwargs,
+        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:
@@ -74,40 +80,73 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
             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
 
-        model_description = self._get_model_description(model_name)
+        # 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
+        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]
+        else:
+            self.device_id = None
+
+        self.model_description = self._get_model_description(model_name)
         self.cache_dir = define_cache_dir(cache_dir)
 
-        model_dir = self.download_model(
-            model_description, self.cache_dir, local_files_only=self._local_files_only
-        )
-
-        self.load_onnx_model(
-            model_dir=model_dir,
-            model_file=model_description["model_file"],
-            threads=threads,
-            providers=providers,
+        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 = {}
 
-        for token, idx in self.tokenizer.get_vocab().items():
-            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.special_tokens = set()
+        self.special_tokens_ids = set()
         self.punctuation = set(string.punctuation)
-        self.stopwords = set(self._load_stopwords(model_dir))
-        self.stemmer = get_stemmer(MODEL_TO_LANGUAGE[model_name])
+        self.stopwords = set(self._load_stopwords(self._model_dir))
+        self.stemmer = SnowballStemmer(MODEL_TO_LANGUAGE[model_name])
         self.alpha = alpha
 
-    def _filter_pair_tokens(self, tokens: List[Tuple[str, Any]]) -> List[Tuple[str, Any]]:
+        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():
+            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 = []
         for token, value in tokens:
             if token in self.stopwords or token in self.punctuation:
@@ -115,17 +154,17 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
             result.append((token, value))
         return result
 
-    def _stem_pair_tokens(self, tokens: List[Tuple[str, Any]]) -> List[Tuple[str, Any]]:
+    def _stem_pair_tokens(self, tokens: list[tuple[str, Any]]) -> list[tuple[str, Any]]:
         result = []
         for token, value in tokens:
-            processed_token = self.stemmer.stemWord(token)
+            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]]:
+        cls, tokens: list[tuple[str, list[int]]], weights: list[float]
+    ) -> list[tuple[str, float]]:
         result = []
         for token, idxs in tokens:
             sum_weight = sum(weights[idx] for idx in idxs)
@@ -133,8 +172,8 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
         return result
 
     def _reconstruct_bpe(
-        self, bpe_tokens: Iterable[Tuple[int, str]]
-    ) -> List[Tuple[str, List[int]]]:
+        self, bpe_tokens: Iterable[tuple[int, str]]
+    ) -> list[tuple[str, list[int]]]:
         result = []
         acc = ""
         acc_idx = []
@@ -161,7 +200,7 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
 
         return result
 
-    def _rescore_vector(self, vector: Dict[str, float]) -> Dict[int, float]:
+    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.
@@ -180,6 +219,9 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
         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
 
         # attention_value shape: (batch_size, num_heads, num_tokens, num_tokens)
@@ -209,16 +251,16 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
             yield SparseEmbedding.from_dict(rescored)
 
     @classmethod
-    def list_supported_models(cls) -> List[Dict[str, Any]]:
+    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.
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
         """
         return supported_bm42_models
 
     @classmethod
-    def _load_stopwords(cls, model_dir: Path) -> List[str]:
+    def _load_stopwords(cls, model_dir: Path) -> list[str]:
         stopwords_path = model_dir / "stopwords.txt"
         if not stopwords_path.exists():
             return []
@@ -231,7 +273,7 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
         documents: Union[str, Iterable[str]],
         batch_size: int = 256,
         parallel: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ) -> Iterable[SparseEmbedding]:
         """
         Encode a list of documents into list of embeddings.
@@ -254,18 +296,23 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
             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]:
+    def _query_rehash(cls, tokens: Iterable[str]) -> dict[int, float]:
         result = {}
         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) -> Iterable[SparseEmbedding]:
+    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.
@@ -274,6 +321,9 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
         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)
             document_tokens_with_ids = enumerate(encoded.tokens)
@@ -289,5 +339,9 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
 
 
 class Bm42TextEmbeddingWorker(TextEmbeddingWorker):
-    def init_embedding(self, model_name: str, cache_dir: str, **kwargs) -> Bm42:
-        return Bm42(model_name=model_name, cache_dir=cache_dir, **kwargs)
+    def init_embedding(self, model_name: str, cache_dir: str, **kwargs: Any) -> Bm42:
+        return Bm42(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            **kwargs,
+        )

fastembed/sparse/sparse_embedding_base.py

@@ -1,5 +1,5 @@
 from dataclasses import dataclass
-from typing import Dict, Iterable, Optional, Union
+from typing import Iterable, Optional, Union, Any
 
 import numpy as np
 
@@ -11,17 +11,17 @@ class SparseEmbedding:
     values: np.ndarray
     indices: np.ndarray
 
-    def as_object(self) -> Dict[str, np.ndarray]:
+    def as_object(self) -> dict[str, np.ndarray]:
         return {
             "values": self.values,
             "indices": self.indices,
         }
 
-    def as_dict(self) -> Dict[int, float]:
+    def as_dict(self) -> dict[int, float]:
         return {i: v for i, v in zip(self.indices, self.values)}
 
     @classmethod
-    def from_dict(cls, data: Dict[int, float]) -> "SparseEmbedding":
+    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())
@@ -34,7 +34,7 @@ class SparseTextEmbeddingBase(ModelManagement):
         model_name: str,
         cache_dir: Optional[str] = None,
         threads: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ):
         self.model_name = model_name
         self.cache_dir = cache_dir
@@ -46,13 +46,11 @@ class SparseTextEmbeddingBase(ModelManagement):
         documents: Union[str, Iterable[str]],
         batch_size: int = 256,
         parallel: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ) -> Iterable[SparseEmbedding]:
         raise NotImplementedError()
 
-    def passage_embed(
-        self, texts: Iterable[str], **kwargs
-    ) -> Iterable[SparseEmbedding]:
+    def passage_embed(self, texts: Iterable[str], **kwargs: Any) -> Iterable[SparseEmbedding]:
         """
         Embeds a list of text passages into a list of embeddings.
 
@@ -68,7 +66,7 @@ class SparseTextEmbeddingBase(ModelManagement):
         yield from self.embed(texts, **kwargs)
 
     def query_embed(
-        self, query: Union[str, Iterable[str]], **kwargs
+        self, query: Union[str, Iterable[str]], **kwargs: Any
     ) -> Iterable[SparseEmbedding]:
         """
         Embeds queries

fastembed/sparse/sparse_text_embedding.py

@@ -1,4 +1,4 @@
-from typing import Any, Dict, Iterable, List, Optional, Sequence, Type, Union
+from typing import Any, Iterable, Optional, Sequence, Type, Union
 
 from fastembed.common import OnnxProvider
 from fastembed.sparse.bm25 import Bm25
@@ -12,15 +12,15 @@ import warnings
 
 
 class SparseTextEmbedding(SparseTextEmbeddingBase):
-    EMBEDDINGS_REGISTRY: List[Type[SparseTextEmbeddingBase]] = [SpladePP, Bm42, Bm25]
+    EMBEDDINGS_REGISTRY: list[Type[SparseTextEmbeddingBase]] = [SpladePP, Bm42, Bm25]
 
     @classmethod
-    def list_supported_models(cls) -> List[Dict[str, Any]]:
+    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.
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
 
             Example:
                 ```
@@ -29,6 +29,7 @@ class SparseTextEmbedding(SparseTextEmbeddingBase):
                         "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",
@@ -48,7 +49,10 @@ class SparseTextEmbedding(SparseTextEmbeddingBase):
         cache_dir: Optional[str] = None,
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
-        **kwargs,
+        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":
@@ -68,6 +72,9 @@ class SparseTextEmbedding(SparseTextEmbeddingBase):
                     cache_dir,
                     threads=threads,
                     providers=providers,
+                    cuda=cuda,
+                    device_ids=device_ids,
+                    lazy_load=lazy_load,
                     **kwargs,
                 )
                 return
@@ -82,7 +89,7 @@ class SparseTextEmbedding(SparseTextEmbeddingBase):
         documents: Union[str, Iterable[str]],
         batch_size: int = 256,
         parallel: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ) -> Iterable[SparseEmbedding]:
         """
         Encode a list of documents into list of embeddings.
@@ -101,7 +108,9 @@ class SparseTextEmbedding(SparseTextEmbeddingBase):
         """
         yield from self.model.embed(documents, batch_size, parallel, **kwargs)
 
-    def query_embed(self, query: Union[str, Iterable[str]], **kwargs) -> Iterable[SparseEmbedding]:
+    def query_embed(
+        self, query: Union[str, Iterable[str]], **kwargs: Any
+    ) -> Iterable[SparseEmbedding]:
         """
         Embeds queries
 

fastembed/sparse/splade_pp.py

@@ -1,4 +1,4 @@
-from typing import Any, Dict, Iterable, List, Optional, Sequence, Type, Union
+from typing import Any, Iterable, Optional, Sequence, Type, Union
 
 import numpy as np
 from fastembed.common import OnnxProvider
@@ -14,7 +14,8 @@ supported_splade_models = [
     {
         "model": "prithivida/Splade_PP_en_v1",
         "vocab_size": 30522,
-        "description": "Independent Implementation of SPLADE++ Model for English",
+        "description": "Independent Implementation of SPLADE++ Model for English.",
+        "license": "apache-2.0",
         "size_in_GB": 0.532,
         "sources": {
             "hf": "Qdrant/SPLADE_PP_en_v1",
@@ -24,7 +25,8 @@ supported_splade_models = [
     {
         "model": "prithvida/Splade_PP_en_v1",
         "vocab_size": 30522,
-        "description": "Independent Implementation of SPLADE++ Model for English",
+        "description": "Independent Implementation of SPLADE++ Model for English.",
+        "license": "apache-2.0",
         "size_in_GB": 0.532,
         "sources": {
             "hf": "Qdrant/SPLADE_PP_en_v1",
@@ -36,6 +38,9 @@ supported_splade_models = [
 
 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)
@@ -50,11 +55,11 @@ class SpladePP(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
             yield SparseEmbedding(values=scores, indices=indices)
 
     @classmethod
-    def list_supported_models(cls) -> List[Dict[str, Any]]:
+    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.
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
         """
         return supported_splade_models
 
@@ -64,7 +69,12 @@ class SpladePP(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
         cache_dir: Optional[str] = None,
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
-        **kwargs,
+        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:
@@ -73,24 +83,57 @@ class SpladePP(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
                                        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
 
-        model_description = self._get_model_description(model_name)
+        # 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
+        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]
+        else:
+            self.device_id = None
+
+        self.model_description = self._get_model_description(model_name)
         self.cache_dir = define_cache_dir(cache_dir)
 
-        model_dir = self.download_model(
-            model_description, self.cache_dir, local_files_only=self._local_files_only
+        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.load_onnx_model(
-            model_dir=model_dir,
-            model_file=model_description["model_file"],
-            threads=threads,
-            providers=providers,
+        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(
@@ -98,7 +141,7 @@ class SpladePP(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
         documents: Union[str, Iterable[str]],
         batch_size: int = 256,
         parallel: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ) -> Iterable[SparseEmbedding]:
         """
         Encode a list of documents into list of embeddings.
@@ -121,6 +164,10 @@ class SpladePP(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
             documents=documents,
             batch_size=batch_size,
             parallel=parallel,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_ids=self.device_ids,
+            **kwargs,
         )
 
     @classmethod
@@ -129,5 +176,10 @@ class SpladePP(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
 
 
 class SpladePPEmbeddingWorker(TextEmbeddingWorker):
-    def init_embedding(self, model_name: str, cache_dir: str, **kwargs) -> SpladePP:
-        return SpladePP(model_name=model_name, cache_dir=cache_dir, threads=1, **kwargs)
+    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,
+        )

fastembed/sparse/utils/tokenizer.py

@@ -1,7 +1,15 @@
 # This code is a modified copy of the `NLTKWordTokenizer` class from `NLTK` library.
 
 import re
-from typing import List
+
+
+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:
@@ -68,12 +76,11 @@ class WordTokenizer:
         )
     ]
     CONTRACTIONS3 = [
-        re.compile(pattern)
-        for pattern in (r"(?i) ('t)(?#X)(is)\b", r"(?i) ('t)(?#X)(was)\b")
+        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]:
+    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.'''

fastembed/text/clip_embedding.py

@@ -1,4 +1,4 @@
-from typing import Any, Dict, Iterable, List, Type
+from typing import Any, Iterable, Type
 
 import numpy as np
 
@@ -11,6 +11,7 @@ supported_clip_models = [
         "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": {
             "hf": "Qdrant/clip-ViT-B-32-text",
@@ -26,24 +27,28 @@ class CLIPOnnxEmbedding(OnnxTextEmbedding):
         return CLIPEmbeddingWorker
 
     @classmethod
-    def list_supported_models(cls) -> List[Dict[str, Any]]:
+    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.
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
         """
         return supported_clip_models
 
-    def _post_process_onnx_output(
-        self, output: OnnxOutputContext
-    ) -> Iterable[np.ndarray]:
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[np.ndarray]:
         return output.model_output
 
 
 class CLIPEmbeddingWorker(OnnxTextEmbeddingWorker):
     def init_embedding(
-        self, model_name: str, cache_dir: str, **kwargs
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
     ) -> OnnxTextEmbedding:
         return CLIPOnnxEmbedding(
-            model_name=model_name, cache_dir=cache_dir, threads=1, **kwargs
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
         )

fastembed/text/e5_onnx_embedding.py

@@ -1,64 +0,0 @@
-from typing import Any, Dict, List, Type
-
-import numpy as np
-
-from fastembed.text.onnx_embedding import OnnxTextEmbedding, OnnxTextEmbeddingWorker
-from fastembed.text.onnx_text_model import TextEmbeddingWorker
-
-supported_multilingual_e5_models = [
-    {
-        "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",
-        "size_in_GB": 2.24,
-        "sources": {
-            "url": "https://storage.googleapis.com/qdrant-fastembed/fast-multilingual-e5-large.tar.gz",
-            "hf": "qdrant/multilingual-e5-large-onnx",
-        },
-        "model_file": "model.onnx",
-        "additional_files": ["model.onnx_data"],
-    },
-    {
-        "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",
-        "size_in_GB": 1.00,
-        "sources": {
-            "hf": "xenova/paraphrase-multilingual-mpnet-base-v2",
-        },
-        "model_file": "onnx/model.onnx",
-    },
-]
-
-
-class E5OnnxEmbedding(OnnxTextEmbedding):
-    @classmethod
-    def _get_worker_class(cls) -> Type["TextEmbeddingWorker"]:
-        return E5OnnxEmbeddingWorker
-
-    @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 supported_multilingual_e5_models
-
-    def _preprocess_onnx_input(
-        self, onnx_input: Dict[str, np.ndarray], **kwargs
-    ) -> Dict[str, np.ndarray]:
-        """
-        Preprocess the onnx input.
-        """
-        onnx_input.pop("token_type_ids", None)
-        return onnx_input
-
-
-class E5OnnxEmbeddingWorker(OnnxTextEmbeddingWorker):
-    def init_embedding(
-        self, model_name: str, cache_dir: str, **kwargs
-    ) -> E5OnnxEmbedding:
-        return E5OnnxEmbedding(
-            model_name=model_name, cache_dir=cache_dir, threads=1, **kwargs
-        )

fastembed/text/multitask_embedding.py

@@ -0,0 +1,98 @@
+from enum import Enum
+from typing import Any, Type, Iterable, Union, Optional
+
+import numpy as np
+
+from fastembed.text.pooled_normalized_embedding import PooledNormalizedEmbedding
+from fastembed.text.onnx_embedding import OnnxTextEmbeddingWorker
+from fastembed.text.onnx_text_model import TextEmbeddingWorker
+
+supported_multitask_models = [
+    {
+        "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": {
+            "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, **kwargs: Any):
+        super().__init__(*args, **kwargs)
+        self._current_task_id = self.PASSAGE_TASK
+
+    @classmethod
+    def _get_worker_class(cls) -> Type["TextEmbeddingWorker"]:
+        return JinaEmbeddingV3Worker
+
+    @classmethod
+    def list_supported_models(cls) -> list[dict[str, Any]]:
+        return supported_multitask_models
+
+    def _preprocess_onnx_input(
+        self, onnx_input: dict[str, np.ndarray], **kwargs
+    ) -> dict[str, np.ndarray]:
+        onnx_input["task_id"] = np.array(self._current_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: int = PASSAGE_TASK,
+        **kwargs,
+    ) -> Iterable[np.ndarray]:
+        self._current_task_id = task_id
+        kwargs["task_id"] = task_id
+        yield from super().embed(documents, batch_size, parallel, **kwargs)
+
+    def query_embed(self, query: Union[str, Iterable[str]], **kwargs) -> Iterable[np.ndarray]:
+        self._current_task_id = self.QUERY_TASK
+        yield from super().embed(query, **kwargs)
+
+    def passage_embed(self, texts: Iterable[str], **kwargs) -> Iterable[np.ndarray]:
+        self._current_task_id = self.PASSAGE_TASK
+        yield from super().embed(texts, **kwargs)
+
+
+class JinaEmbeddingV3Worker(OnnxTextEmbeddingWorker):
+    def init_embedding(
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs,
+    ) -> JinaEmbeddingV3:
+        model = JinaEmbeddingV3(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )
+        model._current_task_id = kwargs["task_id"]
+        return model

fastembed/text/onnx_embedding.py

@@ -1,4 +1,4 @@
-from typing import Any, Dict, Iterable, List, Optional, Sequence, Type, Union
+from typing import Any, Iterable, Optional, Sequence, Type, Union
 
 import numpy as np
 
@@ -12,9 +12,11 @@ supported_onnx_models = [
     {
         "model": "BAAI/bge-base-en",
         "dim": 768,
-        "description": "Text embeddings, Unimodal (text), English, 512 input tokens truncation, Prefixes for queries/documents: necessary, 2023 year",
+        "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": {
+            "hf": "Qdrant/fast-bge-base-en",
             "url": "https://storage.googleapis.com/qdrant-fastembed/fast-bge-base-en.tar.gz",
         },
         "model_file": "model_optimized.onnx",
@@ -22,7 +24,8 @@ supported_onnx_models = [
     {
         "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",
+        "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": {
             "url": "https://storage.googleapis.com/qdrant-fastembed/fast-bge-base-en-v1.5.tar.gz",
@@ -33,7 +36,8 @@ supported_onnx_models = [
     {
         "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",
+        "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": {
             "hf": "qdrant/bge-large-en-v1.5-onnx",
@@ -43,9 +47,11 @@ supported_onnx_models = [
     {
         "model": "BAAI/bge-small-en",
         "dim": 384,
-        "description": "Text embeddings, Unimodal (text), English, 512 input tokens truncation, Prefixes for queries/documents: necessary, 2023 year",
+        "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": {
+            "hf": "Qdrant/bge-small-en",
             "url": "https://storage.googleapis.com/qdrant-fastembed/BAAI-bge-small-en.tar.gz",
         },
         "model_file": "model_optimized.onnx",
@@ -53,7 +59,8 @@ supported_onnx_models = [
     {
         "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",
+        "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": {
             "hf": "qdrant/bge-small-en-v1.5-onnx-q",
@@ -63,27 +70,20 @@ supported_onnx_models = [
     {
         "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",
+        "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": {
+            "hf": "Qdrant/bge-small-zh-v1.5",
             "url": "https://storage.googleapis.com/qdrant-fastembed/fast-bge-small-zh-v1.5.tar.gz",
         },
         "model_file": "model_optimized.onnx",
     },
-    {
-        "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",
-        "size_in_GB": 0.22,
-        "sources": {
-            "hf": "qdrant/paraphrase-multilingual-MiniLM-L12-v2-onnx-Q",
-        },
-        "model_file": "model_optimized.onnx",
-    },
     {
         "model": "thenlper/gte-large",
         "dim": 1024,
-        "description": "Text embeddings, Unimodal (text), English, 512 input tokens truncation, Prefixes for queries/documents: not necessary, 2023 year",
+        "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": {
             "hf": "qdrant/gte-large-onnx",
@@ -93,7 +93,8 @@ supported_onnx_models = [
     {
         "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",
+        "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": {
             "hf": "mixedbread-ai/mxbai-embed-large-v1",
@@ -103,7 +104,8 @@ supported_onnx_models = [
     {
         "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",
+        "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": {
             "hf": "snowflake/snowflake-arctic-embed-xs",
@@ -113,7 +115,8 @@ supported_onnx_models = [
     {
         "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",
+        "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": {
             "hf": "snowflake/snowflake-arctic-embed-s",
@@ -123,7 +126,8 @@ supported_onnx_models = [
     {
         "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",
+        "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": {
             "hf": "Snowflake/snowflake-arctic-embed-m",
@@ -133,7 +137,8 @@ supported_onnx_models = [
     {
         "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",
+        "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": {
             "hf": "snowflake/snowflake-arctic-embed-m-long",
@@ -143,13 +148,25 @@ supported_onnx_models = [
     {
         "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",
+        "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": {
             "hf": "snowflake/snowflake-arctic-embed-l",
         },
         "model_file": "onnx/model.onnx",
     },
+    {
+        "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": {
+            "hf": "jinaai/jina-clip-v1",
+        },
+        "model_file": "onnx/text_model.onnx",
+    },
 ]
 
 
@@ -157,12 +174,12 @@ class OnnxTextEmbedding(TextEmbeddingBase, OnnxTextModel[np.ndarray]):
     """Implementation of the Flag Embedding model."""
 
     @classmethod
-    def list_supported_models(cls) -> List[Dict[str, Any]]:
+    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.
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
         """
         return supported_onnx_models
 
@@ -172,7 +189,12 @@ class OnnxTextEmbedding(TextEmbeddingBase, OnnxTextModel[np.ndarray]):
         cache_dir: Optional[str] = None,
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
-        **kwargs,
+        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:
@@ -181,32 +203,54 @@ class OnnxTextEmbedding(TextEmbeddingBase, OnnxTextModel[np.ndarray]):
                                        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
 
-        model_description = self._get_model_description(model_name)
+        # 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
+        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]
+        else:
+            self.device_id = None
+
+        self.model_description = self._get_model_description(model_name)
         self.cache_dir = define_cache_dir(cache_dir)
-        model_dir = self.download_model(
-            model_description, self.cache_dir, local_files_only=self._local_files_only
+        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.load_onnx_model(
-            model_dir=model_dir,
-            model_file=model_description["model_file"],
-            threads=threads,
-            providers=providers,
-        )
+        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,
+        **kwargs: Any,
     ) -> Iterable[np.ndarray]:
         """
         Encode a list of documents into list of embeddings.
@@ -229,6 +273,9 @@ class OnnxTextEmbedding(TextEmbeddingBase, OnnxTextModel[np.ndarray]):
             documents=documents,
             batch_size=batch_size,
             parallel=parallel,
+            providers=self.providers,
+            cuda=self.cuda,
+            device_ids=self.device_ids,
             **kwargs,
         )
 
@@ -237,8 +284,8 @@ class OnnxTextEmbedding(TextEmbeddingBase, OnnxTextModel[np.ndarray]):
         return OnnxTextEmbeddingWorker
 
     def _preprocess_onnx_input(
-        self, onnx_input: Dict[str, np.ndarray], **kwargs
-    ) -> Dict[str, np.ndarray]:
+        self, onnx_input: dict[str, np.ndarray], **kwargs: Any
+    ) -> dict[str, np.ndarray]:
         """
         Preprocess the onnx input.
         """
@@ -246,7 +293,23 @@ class OnnxTextEmbedding(TextEmbeddingBase, OnnxTextModel[np.ndarray]):
 
     def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[np.ndarray]:
         embeddings = output.model_output
-        return normalize(embeddings[:, 0]).astype(np.float32)
+        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).astype(np.float32)
+
+    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):
@@ -254,6 +317,11 @@ class OnnxTextEmbeddingWorker(TextEmbeddingWorker):
         self,
         model_name: str,
         cache_dir: str,
-        **kwargs,
+        **kwargs: Any,
     ) -> OnnxTextEmbedding:
-        return OnnxTextEmbedding(model_name=model_name, cache_dir=cache_dir, threads=1, **kwargs)
+        return OnnxTextEmbedding(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )

fastembed/text/onnx_text_model.py

@@ -1,7 +1,7 @@
 import os
 from multiprocessing import get_all_start_methods
 from pathlib import Path
-from typing import Any, Dict, Iterable, List, Optional, Sequence, Tuple, Type, Union
+from typing import Any, Iterable, Optional, Sequence, Type, Union
 
 import numpy as np
 from tokenizers import Encoding
@@ -14,7 +14,7 @@ from fastembed.parallel_processor import ParallelWorkerPool
 
 
 class OnnxTextModel(OnnxModel[T]):
-    ONNX_OUTPUT_NAMES: Optional[List[str]] = None
+    ONNX_OUTPUT_NAMES: Optional[list[str]] = None
 
     @classmethod
     def _get_worker_class(cls) -> Type["TextEmbeddingWorker"]:
@@ -23,41 +23,49 @@ class OnnxTextModel(OnnxModel[T]):
     def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[T]:
         raise NotImplementedError("Subclasses must implement this method")
 
-    def __init__(self) -> None:
+    def __init__(self):
         super().__init__()
         self.tokenizer = None
         self.special_token_to_id = {}
 
     def _preprocess_onnx_input(
-        self, onnx_input: Dict[str, np.ndarray], **kwargs
-    ) -> Dict[str, np.ndarray]:
+        self, onnx_input: dict[str, np.ndarray], **kwargs: Any
+    ) -> dict[str, np.ndarray]:
         """
         Preprocess the onnx input.
         """
         return onnx_input
 
-    def load_onnx_model(
+    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,
+        specific_model_path: Optional[str] = None,
     ) -> None:
-        super().load_onnx_model(
+        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 tokenize(self, documents: List[str], **kwargs) -> List[Encoding]:
+    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)
 
     def onnx_embed(
         self,
-        documents: List[str],
-        **kwargs,
+        documents: list[str],
+        **kwargs: Any,
     ) -> OnnxOutputContext:
         encoded = self.tokenize(documents, **kwargs)
         input_ids = np.array([e.ids for e in encoded])
@@ -72,7 +80,6 @@ class OnnxTextModel(OnnxModel[T]):
             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)
@@ -89,7 +96,10 @@ class OnnxTextModel(OnnxModel[T]):
         documents: Union[str, Iterable[str]],
         batch_size: int = 256,
         parallel: Optional[int] = None,
-        **kwargs,
+        providers: Optional[Sequence[OnnxProvider]] = None,
+        cuda: bool = False,
+        device_ids: Optional[list[int]] = None,
+        **kwargs: Any,
     ) -> Iterable[T]:
         is_small = False
 
@@ -101,26 +111,36 @@ class OnnxTextModel(OnnxModel[T]):
             if len(documents) < batch_size:
                 is_small = True
 
-        if parallel == 0:
-            parallel = os.cpu_count()
-
         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))
         else:
-            start_method = (
-                "forkserver" if "forkserver" in get_all_start_methods() else "spawn"
-            )
-            params = {"model_name": model_name, "cache_dir": cache_dir, **kwargs}
+            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(
-                parallel, self._get_worker_class(), start_method=start_method
+                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)
 
 
 class TextEmbeddingWorker(EmbeddingWorker):
-    def process(self, items: Iterable[Tuple[int, Any]]) -> Iterable[Tuple[int, Any]]:
+    def process(self, items: Iterable[tuple[int, Any]]) -> Iterable[tuple[int, Any]]:
         for idx, batch in items:
             onnx_output = self.model.onnx_embed(batch)
             yield idx, onnx_output

fastembed/text/pooled_embedding.py

@@ -1,9 +1,8 @@
-from typing import Any, Dict, Iterable, List, Type
+from typing import Any, Iterable, Type
 
 import numpy as np
 
 from fastembed.common.onnx_model import OnnxOutputContext
-from fastembed.common.utils import normalize
 from fastembed.text.onnx_embedding import OnnxTextEmbedding, OnnxTextEmbeddingWorker
 from fastembed.text.onnx_text_model import TextEmbeddingWorker
 
@@ -11,7 +10,8 @@ supported_pooled_models = [
     {
         "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",
+        "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": {
             "hf": "nomic-ai/nomic-embed-text-v1.5",
@@ -21,7 +21,8 @@ supported_pooled_models = [
     {
         "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",
+        "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": {
             "hf": "nomic-ai/nomic-embed-text-v1.5",
@@ -31,13 +32,49 @@ supported_pooled_models = [
     {
         "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",
+        "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": {
             "hf": "nomic-ai/nomic-embed-text-v1",
         },
         "model_file": "onnx/model.onnx",
     },
+    {
+        "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": {
+            "hf": "qdrant/paraphrase-multilingual-MiniLM-L12-v2-onnx-Q",
+        },
+        "model_file": "model_optimized.onnx",
+    },
+    {
+        "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": {
+            "hf": "xenova/paraphrase-multilingual-mpnet-base-v2",
+        },
+        "model_file": "onnx/model.onnx",
+    },
+    {
+        "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": {
+            "url": "https://storage.googleapis.com/qdrant-fastembed/fast-multilingual-e5-large.tar.gz",
+            "hf": "qdrant/multilingual-e5-large-onnx",
+        },
+        "model_file": "model.onnx",
+        "additional_files": ["model.onnx_data"],
+    },
 ]
 
 
@@ -47,14 +84,10 @@ class PooledEmbedding(OnnxTextEmbedding):
         return PooledEmbeddingWorker
 
     @classmethod
-    def mean_pooling(
-        cls, model_output: np.ndarray, attention_mask: np.ndarray
-    ) -> np.ndarray:
+    def mean_pooling(cls, model_output: np.ndarray, attention_mask: np.ndarray) -> np.ndarray:
         token_embeddings = model_output
         input_mask_expanded = np.expand_dims(attention_mask, axis=-1)
-        input_mask_expanded = np.tile(
-            input_mask_expanded, (1, 1, token_embeddings.shape[-1])
-        )
+        input_mask_expanded = np.tile(input_mask_expanded, (1, 1, token_embeddings.shape[-1]))
         input_mask_expanded = input_mask_expanded.astype(float)
         sum_embeddings = np.sum(token_embeddings * input_mask_expanded, axis=1)
         sum_mask = np.sum(input_mask_expanded, axis=1)
@@ -62,17 +95,18 @@ class PooledEmbedding(OnnxTextEmbedding):
         return pooled_embeddings
 
     @classmethod
-    def list_supported_models(cls) -> List[Dict[str, Any]]:
+    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.
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
         """
         return supported_pooled_models
 
-    def _post_process_onnx_output(
-        self, output: OnnxOutputContext
-    ) -> Iterable[np.ndarray]:
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[np.ndarray]:
+        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).astype(np.float32)
@@ -80,8 +114,14 @@ class PooledEmbedding(OnnxTextEmbedding):
 
 class PooledEmbeddingWorker(OnnxTextEmbeddingWorker):
     def init_embedding(
-        self, model_name: str, cache_dir: str, **kwargs
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
     ) -> OnnxTextEmbedding:
         return PooledEmbedding(
-            model_name=model_name, cache_dir=cache_dir, threads=1, **kwargs
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
         )

fastembed/text/pooled_normalized_embedding.py

@@ -1,4 +1,4 @@
-from typing import Any, Dict, Iterable, List, Type
+from typing import Any, Iterable, Type
 
 import numpy as np
 
@@ -12,7 +12,8 @@ supported_pooled_normalized_models = [
     {
         "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",
+        "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": {
             "url": "https://storage.googleapis.com/qdrant-fastembed/sentence-transformers-all-MiniLM-L6-v2.tar.gz",
@@ -23,7 +24,8 @@ supported_pooled_normalized_models = [
     {
         "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",
+        "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": {"hf": "xenova/jina-embeddings-v2-base-en"},
         "model_file": "onnx/model.onnx",
@@ -31,7 +33,8 @@ supported_pooled_normalized_models = [
     {
         "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",
+        "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": {"hf": "xenova/jina-embeddings-v2-small-en"},
         "model_file": "onnx/model.onnx",
@@ -39,7 +42,8 @@ supported_pooled_normalized_models = [
     {
         "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",
+        "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": {"hf": "jinaai/jina-embeddings-v2-base-de"},
         "model_file": "onnx/model_fp16.onnx",
@@ -47,11 +51,39 @@ supported_pooled_normalized_models = [
     {
         "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",
+        "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": {"hf": "jinaai/jina-embeddings-v2-base-code"},
         "model_file": "onnx/model.onnx",
     },
+    {
+        "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": {"hf": "jinaai/jina-embeddings-v2-base-zh"},
+        "model_file": "onnx/model.onnx",
+    },
+    {
+        "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": {"hf": "jinaai/jina-embeddings-v2-base-es"},
+        "model_file": "onnx/model.onnx",
+    },
+    {
+        "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": {"hf": "thenlper/gte-base"},
+        "model_file": "onnx/model.onnx",
+    },
 ]
 
 
@@ -61,17 +93,18 @@ class PooledNormalizedEmbedding(PooledEmbedding):
         return PooledNormalizedEmbeddingWorker
 
     @classmethod
-    def list_supported_models(cls) -> List[Dict[str, Any]]:
+    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.
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
         """
         return supported_pooled_normalized_models
 
-    def _post_process_onnx_output(
-        self, output: OnnxOutputContext
-    ) -> Iterable[np.ndarray]:
+    def _post_process_onnx_output(self, output: OnnxOutputContext) -> Iterable[np.ndarray]:
+        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)).astype(np.float32)
@@ -79,8 +112,14 @@ class PooledNormalizedEmbedding(PooledEmbedding):
 
 class PooledNormalizedEmbeddingWorker(OnnxTextEmbeddingWorker):
     def init_embedding(
-        self, model_name: str, cache_dir: str, **kwargs
+        self,
+        model_name: str,
+        cache_dir: str,
+        **kwargs: Any,
     ) -> OnnxTextEmbedding:
         return PooledNormalizedEmbedding(
-            model_name=model_name, cache_dir=cache_dir, threads=1, **kwargs
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
         )

fastembed/text/text_embedding.py

@@ -1,32 +1,32 @@
-from typing import Any, Dict, Iterable, List, Optional, Sequence, Type, Union
+import warnings
+from typing import Any, Iterable, Optional, Sequence, Type, Union
 
 import numpy as np
-
 from fastembed.common import OnnxProvider
 from fastembed.text.clip_embedding import CLIPOnnxEmbedding
-from fastembed.text.e5_onnx_embedding import E5OnnxEmbedding
 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
 
 
 class TextEmbedding(TextEmbeddingBase):
-    EMBEDDINGS_REGISTRY: List[Type[TextEmbeddingBase]] = [
+    EMBEDDINGS_REGISTRY: list[Type[TextEmbeddingBase]] = [
         OnnxTextEmbedding,
-        E5OnnxEmbedding,
         CLIPOnnxEmbedding,
         PooledNormalizedEmbedding,
         PooledEmbedding,
+        JinaEmbeddingV3,
     ]
 
     @classmethod
-    def list_supported_models(cls) -> List[Dict[str, Any]]:
+    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.
+            list[dict[str, Any]]: A list of dictionaries containing the model information.
 
             Example:
                 ```
@@ -35,6 +35,7 @@ class TextEmbedding(TextEmbeddingBase):
                         "model": "intfloat/multilingual-e5-large",
                         "dim": 1024,
                         "description": "Multilingual model, e5-large. Recommend using this model for non-English languages",
+                        "license": "mit",
                         "size_in_GB": 2.24,
                         "sources": {
                             "gcp": "https://storage.googleapis.com/qdrant-fastembed/fast-multilingual-e5-large.tar.gz",
@@ -55,27 +56,54 @@ class TextEmbedding(TextEmbeddingBase):
         cache_dir: Optional[str] = None,
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
-        **kwargs,
+        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 and release notes to ensure compatibility with your workflow. ",
+                UserWarning,
+                stacklevel=2,
+            )
+        if model_name == "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2":
+            warnings.warn(
+                "The model 'sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2' has been updated to "
+                "include a mean pooling layer. Please ensure your usage aligns with the new functionality. "
+                "Support for the previous version without mean pooling will be removed as of version 0.5.2.",
+                UserWarning,
+                stacklevel=2,
+            )
+        if model_name in {
+            "sentence-transformers/paraphrase-multilingual-mpnet-base-v2",
+            "intfloat/multilingual-e5-large",
+        }:
+            warnings.warn(
+                f"{model_name} has been updated as of fastembed 0.5.2, outputs are now average pooled.",
+                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
-            ):
+            if any(model_name.lower() == model["model"].lower() for model in supported_models):
                 self.model = EMBEDDING_MODEL_TYPE(
-                    model_name,
-                    cache_dir,
+                    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."
+            f"Model {model_name} is not supported in TextEmbedding. "
             "Please check the supported models using `TextEmbedding.list_supported_models()`"
         )
 
@@ -84,7 +112,7 @@ class TextEmbedding(TextEmbeddingBase):
         documents: Union[str, Iterable[str]],
         batch_size: int = 256,
         parallel: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ) -> Iterable[np.ndarray]:
         """
         Encode a list of documents into list of embeddings.
@@ -102,3 +130,30 @@ class TextEmbedding(TextEmbeddingBase):
             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[np.ndarray]:
+        """
+        Embeds queries
+
+        Args:
+            query (Union[str, Iterable[str]]): The query to embed, or an iterable e.g. list of queries.
+
+        Returns:
+            Iterable[np.ndarray]: 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[np.ndarray]:
+        """
+        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)

fastembed/text/text_embedding_base.py

@@ -1,4 +1,4 @@
-from typing import Iterable, Optional, Union
+from typing import Iterable, Optional, Union, Any
 
 import numpy as np
 
@@ -11,7 +11,7 @@ class TextEmbeddingBase(ModelManagement):
         model_name: str,
         cache_dir: Optional[str] = None,
         threads: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ):
         self.model_name = model_name
         self.cache_dir = cache_dir
@@ -23,11 +23,11 @@ class TextEmbeddingBase(ModelManagement):
         documents: Union[str, Iterable[str]],
         batch_size: int = 256,
         parallel: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ) -> Iterable[np.ndarray]:
         raise NotImplementedError()
 
-    def passage_embed(self, texts: Iterable[str], **kwargs) -> Iterable[np.ndarray]:
+    def passage_embed(self, texts: Iterable[str], **kwargs: Any) -> Iterable[np.ndarray]:
         """
         Embeds a list of text passages into a list of embeddings.
 
@@ -42,9 +42,8 @@ class TextEmbeddingBase(ModelManagement):
         # 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
-    ) -> Iterable[np.ndarray]:
+    def query_embed(self, query: Union[str, Iterable[str]], **kwargs: Any) -> Iterable[np.ndarray]:
+
         """
         Embeds queries
 

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "fastembed"
-version = "0.3.6"
+version = "0.5.1"
 description = "Fast, light, accurate library built for retrieval embedding generation"
 authors = ["Qdrant Team <info@qdrant.tech>", "NirantK <nirant.bits@gmail.com>"]
 license = "Apache License"
@@ -11,28 +11,35 @@ repository = "https://github.com/qdrant/fastembed"
 keywords = ["vector", "embedding", "neural", "search", "qdrant", "sentence-transformers"]
 
 [tool.poetry.dependencies]
-python = ">=3.8.0,<3.13"
-onnx = "^1.15.0"
-onnxruntime = "^1.17.0"
+python = ">=3.9.0"
+numpy = [
+    { version = ">=1.21,<2.1.0", python = "<3.10" },
+    { version = ">=1.21", python = ">=3.10,<3.12" },
+    { version = ">=1.26", python = ">=3.12,<3.13" },
+    { version = ">=2.1.0", python = ">=3.13" }
+]
+onnxruntime = [
+    { version = ">=1.17.0,<1.20.0", python = "<3.10" },
+    { version = ">=1.17.0,!=1.20.0", python = ">=3.10,<3.13" },
+    { version = ">1.20.0", python = ">=3.13" }
+]
 tqdm = "^4.66"
 requests = "^2.31"
 tokenizers = ">=0.15,<1.0"
 huggingface-hub = ">=0.20,<1.0"
 loguru = "^0.7.2"
-numpy = [
-    { version = ">=1.21, <2", python = "<3.12" },
-    { version = ">=1.26, <2", python = ">=3.12" }
-]
 pillow = "^10.3.0"
-snowballstemmer = "^2.2.0"
-PyStemmer = "^2.2.0"
-mmh3 = "^4.0"
+mmh3 = "^4.1.0"
+py-rust-stemmers = "^0.1.0"
 
-[tool.poetry.group.dev.dependencies]
+[tool.poetry.group.test.dependencies]
 pytest = "^7.4.2"
 ruff = ">=0.3.1,<1.0"
+
+[tool.poetry.group.dev.dependencies]
 notebook = ">=7.0.2"
-pre-commit = {version = "^3.6.2", python = ">=3.9,<3.12" }
+pre-commit = "^3.6.2"
+onnx = ">=1.15.0"
 
 [tool.poetry.group.docs.dependencies]
 mkdocs-material = "^9.5.10"
@@ -41,10 +48,16 @@ pillow = "^10.2.0"
 cairosvg = "^2.7.1"
 mknotebooks = "^0.8.0"
 
+[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.pyright]
+typeCheckingMode = "strict"
+
 [tool.ruff]
 line-length = 99

tests/profiling.py

@@ -9,7 +9,7 @@
 
 # %%
 import time
-from typing import Callable, List, Tuple
+from typing import Callable
 
 import matplotlib.pyplot as plt
 import torch.nn.functional as F
@@ -23,7 +23,7 @@ from fastembed.embedding import DefaultEmbedding
 # 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",
@@ -56,7 +56,7 @@ class HF:
         self.model = AutoModel.from_pretrained(model_id)
         self.tokenizer = AutoTokenizer.from_pretrained(model_id)
 
-    def embed(self, texts: List[str]):
+    def embed(self, texts: list[str]):
         encoded_input = self.tokenizer(
             texts, max_length=512, padding=True, truncation=True, return_tensors="pt"
         )
@@ -88,7 +88,7 @@ embedding_model = DefaultEmbedding()
 # %%
 def calculate_time_stats(
     embed_func: Callable, documents: list, k: int
-) -> Tuple[float, float, float]:
+) -> tuple[float, float, float]:
     times = []
     for _ in range(k):
         # Timing the embed_func call
@@ -105,16 +105,14 @@ def calculate_time_stats(
 # %%
 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
-)
+fst_stats = calculate_time_stats(lambda x: list(embedding_model.embed(x)), documents, k=2)
 print(f"FastEmbed (Average, Max, Min): {fst_stats}")
 
 
 # %%
 def plot_character_per_second_comparison(
-    hf_stats: Tuple[float, float, float],
-    fst_stats: Tuple[float, float, float],
+    hf_stats: tuple[float, float, float],
+    fst_stats: tuple[float, float, float],
     documents: list,
 ):
     # Calculating total characters in documents

tests/test_attention_embeddings.py

@@ -1,11 +1,15 @@
+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):
+def test_attention_embeddings(model_name: str) -> None:
+    is_ci = os.getenv("CI")
     model = SparseTextEmbedding(model_name=model_name)
 
     output = list(
@@ -62,9 +66,14 @@ def test_attention_embeddings(model_name):
         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):
+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
@@ -82,9 +91,14 @@ def test_parallel_processing(model_name):
         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):
+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")
@@ -92,13 +106,54 @@ def test_multilanguage(model_name):
     assert embeddings[0].values.shape == (3,)
     assert embeddings[0].indices.shape == (3,)
 
-    assert embeddings[1].values.shape == (2,)
-    assert embeddings[1].indices.shape == (2,)
+    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 == (4,)
-    assert embeddings[0].indices.shape == (4,)
+    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))

tests/test_image_onnx_embeddings.py

@@ -8,6 +8,7 @@ from PIL import Image
 
 from fastembed import ImageEmbedding
 from tests.config import TEST_MISC_DIR
+from tests.utils import delete_model_cache
 
 CANONICAL_VECTOR_VALUES = {
     "Qdrant/clip-ViT-B-32-vision": np.array([-0.0098, 0.0128, -0.0274, 0.002, -0.0059]),
@@ -20,10 +21,13 @@ CANONICAL_VECTOR_VALUES = {
     "Qdrant/Unicom-ViT-B-32": np.array(
         [0.0418, 0.0550, 0.0003, 0.0253, -0.0185, 0.0016, -0.0368, -0.0402, -0.0891, -0.0186]
     ),
+    "jinaai/jina-clip-v1": np.array(
+        [-0.029, 0.0216, 0.0396, 0.0283, -0.0023, 0.0151, 0.011, -0.0235, 0.0251, -0.0343]
+    ),
 }
 
 
-def test_embedding():
+def test_embedding() -> None:
     is_ci = os.getenv("CI")
 
     for model_desc in ImageEmbedding.list_supported_models():
@@ -52,9 +56,13 @@ def test_embedding():
 
         assert np.allclose(embeddings[1], embeddings[2]), model_desc["model"]
 
+        if is_ci:
+            delete_model_cache(model.model._model_dir)
+
 
 @pytest.mark.parametrize("n_dims,model_name", [(512, "Qdrant/clip-ViT-B-32-vision")])
-def test_batch_embedding(n_dims, model_name):
+def test_batch_embedding(n_dims: int, model_name: str) -> None:
+    is_ci = os.getenv("CI")
     model = ImageEmbedding(model_name=model_name)
     n_images = 32
     test_images = [
@@ -68,10 +76,13 @@ def test_batch_embedding(n_dims, model_name):
     embeddings = np.stack(embeddings, axis=0)
 
     assert embeddings.shape == (len(test_images) * n_images, n_dims)
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize("n_dims,model_name", [(512, "Qdrant/clip-ViT-B-32-vision")])
-def test_parallel_processing(n_dims, model_name):
+def test_parallel_processing(n_dims: int, model_name: str) -> None:
+    is_ci = os.getenv("CI")
     model = ImageEmbedding(model_name=model_name)
 
     n_images = 32
@@ -93,3 +104,20 @@ def test_parallel_processing(n_dims, model_name):
     assert embeddings.shape == (n_images * len(test_images), n_dims)
     assert np.allclose(embeddings, embeddings_2, atol=1e-3)
     assert np.allclose(embeddings, embeddings_3, atol=1e-3)
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
+
+
+@pytest.mark.parametrize("model_name", ["Qdrant/clip-ViT-B-32-vision"])
+def test_lazy_load(model_name: str) -> None:
+    is_ci = os.getenv("CI")
+    model = ImageEmbedding(model_name=model_name, lazy_load=True)
+    assert not hasattr(model.model, "model")
+    images = [
+        TEST_MISC_DIR / "image.jpeg",
+        str(TEST_MISC_DIR / "small_image.jpeg"),
+    ]
+    list(model.embed(images))
+    assert hasattr(model.model, "model")
+    if is_ci:
+        delete_model_cache(model.model._model_dir)

tests/test_late_interaction_embeddings.py

@@ -1,8 +1,12 @@
+import os
+
+import pytest
 import numpy as np
 
 from fastembed.late_interaction.late_interaction_text_embedding import (
     LateInteractionTextEmbedding,
 )
+from tests.utils import delete_model_cache
 
 # vectors are abridged and rounded for brevity
 CANONICAL_COLUMN_VALUES = {
@@ -24,6 +28,15 @@ CANONICAL_COLUMN_VALUES = {
             [-0.07281, 0.04633, -0.04711, 0.00762, -0.07374],
         ]
     ),
+    "jinaai/jina-colbert-v2": np.array(
+        [
+            [0.0742, 0.0591, -0.2403, -0.1774, 0.02],
+            [0.1318, 0.0882, -0.1138, -0.2066, 0.146],
+            [-0.0183, -0.1354, -0.0139, -0.1079, -0.051],
+            [0.0003, -0.1184, -0.07, -0.0479, -0.0649],
+            [0.0766, 0.0452, -0.2343, -0.183, 0.0058],
+        ]
+    ),
 }
 
 CANONICAL_QUERY_VALUES = {
@@ -99,12 +112,49 @@ CANONICAL_QUERY_VALUES = {
             [-0.03473, 0.04792, -0.07033, 0.02196, -0.05314],
         ]
     ),
+    "jinaai/jina-colbert-v2": np.array(
+        [
+            [0.0477, 0.0255, -0.2224, -0.1085, -0.03],
+            [0.0206, -0.0845, -0.0075, -0.1712, 0.0156],
+            [-0.0056, -0.0957, -0.0147, -0.1277, -0.0225],
+            [0.0486, -0.0499, -0.1609, 0.0194, 0.0274],
+            [0.0481, 0.0253, -0.2278, -0.1126, -0.0294],
+            [0.0599, -0.0678, -0.0956, -0.0757, 0.0236],
+            [0.0592, -0.0862, -0.0621, -0.1084, 0.0155],
+            [0.0874, -0.0714, -0.0772, -0.1414, 0.037],
+            [0.1009, -0.0552, -0.0669, -0.163, 0.0493],
+            [0.1135, -0.047, -0.0576, -0.1699, 0.0538],
+            [0.1228, -0.0428, -0.0507, -0.1725, 0.0562],
+            [0.1291, -0.0388, -0.042, -0.1753, 0.0569],
+            [0.1365, -0.0337, -0.0326, -0.1786, 0.0574],
+            [0.1439, -0.026, -0.024, -0.1831, 0.0574],
+            [0.1527, -0.0099, -0.0179, -0.1874, 0.057],
+            [0.1555, 0.0186, -0.023, -0.1801, 0.0539],
+            [0.1389, 0.054, -0.0345, -0.1636, 0.0429],
+            [0.1058, 0.0862, -0.0418, -0.1455, 0.0222],
+            [0.0713, 0.1061, -0.0438, -0.1288, 0.0002],
+            [0.0453, 0.1143, -0.0457, -0.1119, -0.019],
+            [0.0346, 0.1131, -0.0487, -0.0952, -0.0338],
+            [0.0355, 0.1073, -0.0493, -0.0823, -0.0438],
+            [0.0424, 0.1041, -0.0459, -0.0761, -0.048],
+            [0.048, 0.102, -0.0421, -0.0718, -0.0477],
+            [0.0474, 0.0989, -0.0413, -0.0654, -0.0431],
+            [0.0434, 0.095, -0.0415, -0.0589, -0.0345],
+            [0.0408, 0.0897, -0.0405, -0.0554, -0.0197],
+            [0.0433, 0.0811, -0.0407, -0.0545, 0.0055],
+            [0.0514, 0.0629, -0.0446, -0.0549, 0.0368],
+            [0.058, 0.048, -0.0527, -0.0607, 0.0568],
+            [0.0561, 0.0447, -0.0661, -0.0702, 0.0764],
+            [0.0204, -0.0856, -0.0386, -0.1232, -0.0332],
+        ]
+    ),
 }
 
 docs = ["Hello World"]
 
 
 def test_batch_embedding():
+    is_ci = os.getenv("CI")
     docs_to_embed = docs * 10
 
     for model_name, expected_result in CANONICAL_COLUMN_VALUES.items():
@@ -114,10 +164,14 @@ def test_batch_embedding():
 
         for value in result:
             token_num, abridged_dim = expected_result.shape
-            assert np.allclose(value[:, :abridged_dim], expected_result, atol=10e-4)
+            assert np.allclose(value[:, :abridged_dim], expected_result, atol=2e-3)
+
+        if is_ci:
+            delete_model_cache(model.model._model_dir)
 
 
 def test_single_embedding():
+    is_ci = os.getenv("CI")
     docs_to_embed = docs
 
     for model_name, expected_result in CANONICAL_COLUMN_VALUES.items():
@@ -125,10 +179,14 @@ def test_single_embedding():
         model = LateInteractionTextEmbedding(model_name=model_name)
         result = next(iter(model.embed(docs_to_embed, batch_size=6)))
         token_num, abridged_dim = expected_result.shape
-        assert np.allclose(result[:, :abridged_dim], expected_result, atol=10e-4)
+        assert np.allclose(result[:, :abridged_dim], expected_result, atol=2e-3)
+
+        if is_ci:
+            delete_model_cache(model.model._model_dir)
 
 
 def test_single_embedding_query():
+    is_ci = os.getenv("CI")
     queries_to_embed = docs
 
     for model_name, expected_result in CANONICAL_QUERY_VALUES.items():
@@ -136,10 +194,14 @@ def test_single_embedding_query():
         model = LateInteractionTextEmbedding(model_name=model_name)
         result = next(iter(model.query_embed(queries_to_embed)))
         token_num, abridged_dim = expected_result.shape
-        assert np.allclose(result[:, :abridged_dim], expected_result, atol=10e-4)
+        assert np.allclose(result[:, :abridged_dim], expected_result, atol=2e-3)
+
+        if is_ci:
+            delete_model_cache(model.model._model_dir)
 
 
 def test_parallel_processing():
+    is_ci = os.getenv("CI")
     model = LateInteractionTextEmbedding(model_name="colbert-ir/colbertv2.0")
     token_dim = 128
     docs = ["hello world", "flag embedding"] * 100
@@ -155,3 +217,30 @@ def test_parallel_processing():
     assert embeddings.shape[0] == len(docs) and embeddings.shape[-1] == token_dim
     assert np.allclose(embeddings, embeddings_2, atol=1e-3)
     assert np.allclose(embeddings, embeddings_3, atol=1e-3)
+
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
+
+
+@pytest.mark.parametrize(
+    "model_name",
+    ["colbert-ir/colbertv2.0"],
+)
+def test_lazy_load(model_name: str):
+    is_ci = os.getenv("CI")
+
+    model = LateInteractionTextEmbedding(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 = LateInteractionTextEmbedding(model_name=model_name, lazy_load=True)
+    list(model.query_embed(docs))
+
+    model = LateInteractionTextEmbedding(model_name=model_name, lazy_load=True)
+    list(model.passage_embed(docs))
+
+    if is_ci:
+        delete_model_cache(model.model._model_dir)

tests/test_multi_gpu.py

@@ -0,0 +1,221 @@
+import pytest
+from typing import Optional
+from fastembed import (
+    TextEmbedding,
+    SparseTextEmbedding,
+    LateInteractionTextEmbedding,
+    ImageEmbedding,
+)
+from fastembed.rerank.cross_encoder import TextCrossEncoder
+from tests.config import TEST_MISC_DIR
+
+CACHE_DIR = "../model_cache"
+
+
+@pytest.mark.skip(reason="Requires a multi-gpu server")
+@pytest.mark.parametrize("device_id", [None, 0, 1])
+def test_gpu_via_providers(device_id: Optional[int]) -> None:
+    docs = ["hello world", "flag embedding"]
+
+    device_id = device_id if device_id is not None else 0
+    providers = (
+        ["CUDAExecutionProvider"]
+        if device_id is None
+        else [("CUDAExecutionProvider", {"device_id": device_id})]
+    )
+    embedding_model = TextEmbedding(
+        "sentence-transformers/all-MiniLM-L6-v2",
+        providers=providers,
+        cache_dir=CACHE_DIR,
+    )
+    list(embedding_model.embed(docs))
+    options = embedding_model.model.model.get_provider_options()
+    assert options["CUDAExecutionProvider"]["device_id"] == str(device_id)
+
+    embedding_model = SparseTextEmbedding(
+        "prithvida/Splade_PP_en_v1",
+        providers=providers,
+        cache_dir=CACHE_DIR,
+    )
+    list(embedding_model.embed(docs))
+    options = embedding_model.model.model.get_provider_options()
+    assert options["CUDAExecutionProvider"]["device_id"] == str(device_id)
+
+    embedding_model = SparseTextEmbedding(
+        "Qdrant/bm42-all-minilm-l6-v2-attentions",
+        providers=providers,
+        cache_dir=CACHE_DIR,
+    )
+    list(embedding_model.embed(docs))
+    options = embedding_model.model.model.get_provider_options()
+    assert options["CUDAExecutionProvider"]["device_id"] == str(device_id)
+
+    embedding_model = LateInteractionTextEmbedding(
+        "colbert-ir/colbertv2.0",
+        providers=providers,
+        cache_dir=CACHE_DIR,
+    )
+    list(embedding_model.embed(docs))
+    options = embedding_model.model.model.get_provider_options()
+    assert options["CUDAExecutionProvider"]["device_id"] == str(device_id)
+
+    embedding_model = ImageEmbedding(
+        model_name="Qdrant/clip-ViT-B-32-vision",
+        providers=providers,
+        cache_dir=CACHE_DIR,
+    )
+    images = [
+        TEST_MISC_DIR / "image.jpeg",
+        str(TEST_MISC_DIR / "small_image.jpeg"),
+    ]
+    list(embedding_model.embed(images))
+    options = embedding_model.model.model.get_provider_options()
+    assert options["CUDAExecutionProvider"]["device_id"] == str(device_id)
+
+    model = TextCrossEncoder(
+        model_name="Xenova/ms-marco-MiniLM-L-6-v2",
+        providers=providers,
+        cache_dir=CACHE_DIR,
+    )
+    query = "What is the capital of France?"
+    documents = ["Paris is the capital of France.", "Berlin is the capital of Germany."]
+    list(model.rerank(query, documents))
+    options = embedding_model.model.model.get_provider_options()
+    assert options["CUDAExecutionProvider"]["device_id"] == str(device_id)
+
+
+@pytest.mark.skip(reason="Requires a multi-gpu server")
+@pytest.mark.parametrize("device_ids", [None, [0], [1], [0, 1]])
+def test_gpu_cuda_device_ids(device_ids: Optional[list[int]]) -> None:
+    docs = ["hello world", "flag embedding"]
+    device_id = device_ids[0] if device_ids else 0
+    embedding_model = TextEmbedding(
+        "sentence-transformers/all-MiniLM-L6-v2",
+        cuda=True,
+        device_ids=device_ids,
+        cache_dir=CACHE_DIR,
+    )
+    list(embedding_model.embed(docs))
+    options = embedding_model.model.model.get_provider_options()
+    assert options["CUDAExecutionProvider"]["device_id"] == str(
+        device_id
+    ), f"Text embedding: {options}"
+
+    embedding_model = SparseTextEmbedding(
+        "prithvida/Splade_PP_en_v1",
+        cuda=True,
+        device_ids=device_ids,
+        cache_dir=CACHE_DIR,
+    )
+    list(embedding_model.embed(docs))
+    options = embedding_model.model.model.get_provider_options()
+    assert options["CUDAExecutionProvider"]["device_id"] == str(
+        device_id
+    ), f"Sparse text embedding: {options}"
+
+    embedding_model = SparseTextEmbedding(
+        "Qdrant/bm42-all-minilm-l6-v2-attentions",
+        cuda=True,
+        device_ids=device_ids,
+        cache_dir=CACHE_DIR,
+    )
+    list(embedding_model.embed(docs))
+    options = embedding_model.model.model.get_provider_options()
+    assert options["CUDAExecutionProvider"]["device_id"] == str(device_id), f"Bm42: {options}"
+
+    embedding_model = LateInteractionTextEmbedding(
+        "colbert-ir/colbertv2.0",
+        cuda=True,
+        device_ids=device_ids,
+        cache_dir=CACHE_DIR,
+    )
+    list(embedding_model.embed(docs))
+    options = embedding_model.model.model.get_provider_options()
+    assert options["CUDAExecutionProvider"]["device_id"] == str(
+        device_id
+    ), f"Late interaction text embedding: {options}"
+
+    embedding_model = ImageEmbedding(
+        model_name="Qdrant/clip-ViT-B-32-vision",
+        cuda=True,
+        device_ids=device_ids,
+        cache_dir=CACHE_DIR,
+    )
+    images = [
+        TEST_MISC_DIR / "image.jpeg",
+        str(TEST_MISC_DIR / "small_image.jpeg"),
+    ]
+    list(embedding_model.embed(images))
+    options = embedding_model.model.model.get_provider_options()
+    assert options["CUDAExecutionProvider"]["device_id"] == str(
+        device_id
+    ), f"Image embedding: {options}"
+
+    if device_ids is None or len(device_ids) == 1:
+        model = TextCrossEncoder(
+            model_name="Xenova/ms-marco-MiniLM-L-6-v2",
+            cuda=True,
+            device_ids=device_ids,
+            cache_dir=CACHE_DIR,
+        )
+        query = "What is the capital of France?"
+        documents = ["Paris is the capital of France.", "Berlin is the capital of Germany."]
+        list(model.rerank(query, documents))
+        options = embedding_model.model.model.get_provider_options()
+        assert options["CUDAExecutionProvider"]["device_id"] == str(
+            device_id
+        ), f"Text cross encoder: {options}"
+
+
+@pytest.mark.skip(reason="Requires a multi-gpu server")
+@pytest.mark.parametrize(
+    "device_ids,parallel", [(None, None), (None, 2), ([1], None), ([1], 1), ([1], 2), ([0, 1], 2)]
+)
+def test_multi_gpu_parallel_inference(device_ids: Optional[list[int]], parallel: int) -> None:
+    docs = ["hello world", "flag embedding"] * 100
+    batch_size = 5
+
+    embedding_model = TextEmbedding(
+        "sentence-transformers/all-MiniLM-L6-v2",
+        cuda=True,
+        device_ids=device_ids,
+        cache_dir=CACHE_DIR,
+        lazy_load=True,
+    )
+    list(embedding_model.embed(docs, batch_size=batch_size, parallel=parallel))
+
+    embedding_model = SparseTextEmbedding(
+        "prithvida/Splade_PP_en_v1",
+        cuda=True,
+        device_ids=device_ids,
+        cache_dir=CACHE_DIR,
+    )
+    list(embedding_model.embed(docs, batch_size=batch_size, parallel=parallel))
+
+    embedding_model = SparseTextEmbedding(
+        "Qdrant/bm42-all-minilm-l6-v2-attentions",
+        cuda=True,
+        device_ids=device_ids,
+        cache_dir=CACHE_DIR,
+    )
+    list(embedding_model.embed(docs, batch_size=batch_size, parallel=parallel))
+
+    embedding_model = LateInteractionTextEmbedding(
+        "colbert-ir/colbertv2.0",
+        cuda=True,
+        device_ids=device_ids,
+        cache_dir=CACHE_DIR,
+    )
+    list(embedding_model.embed(docs, batch_size=batch_size, parallel=parallel))
+
+    embedding_model = ImageEmbedding(
+        model_name="Qdrant/clip-ViT-B-32-vision",
+        cuda=True,
+        device_ids=device_ids,
+        cache_dir=CACHE_DIR,
+    )
+    images = [
+        TEST_MISC_DIR / "image.jpeg",
+        str(TEST_MISC_DIR / "small_image.jpeg"),
+    ] * 100
+    list(embedding_model.embed(images, batch_size=batch_size, parallel=parallel))

tests/test_pooling.py

@@ -0,0 +1,34 @@
+import os
+
+import numpy as np
+
+from fastembed.late_interaction.late_interaction_text_embedding import (
+    LateInteractionTextEmbedding,
+)
+from fastembed.common.pooling import LateInteractionPooler
+
+from tests.utils import delete_model_cache
+
+CANONICAL_COLUMN_VALUES = {
+    "colbert-ir/colbertv2.0": np.array(
+        [4.0727495e-03, -2.4026826e-03, -6.8204990e-04, -7.1383954e-05, 4.4963313e-03]
+    ),
+}
+
+docs = ["Hello World"]
+
+
+def test_batch_embedding():
+    is_ci = os.getenv("CI")
+    docs_to_embed = docs * 10
+
+    for model_name, expected_result in CANONICAL_COLUMN_VALUES.items():
+        print("evaluating", model_name)
+        model = LateInteractionTextEmbedding(model_name=model_name)
+        pooler = LateInteractionPooler()
+        result = list(model.embed(docs_to_embed, batch_size=6))
+        pooled_result = pooler.pool(result)
+        assert np.allclose(pooled_result[0], expected_result, atol=2e-3)
+
+        if is_ci:
+            delete_model_cache(model.model._model_dir)

tests/test_sparse_embeddings.py

@@ -1,6 +1,11 @@
-import pytest
+import os
 
+import pytest
+import numpy as np
+
+from fastembed.sparse.bm25 import Bm25
 from fastembed.sparse.sparse_text_embedding import SparseTextEmbedding
+from tests.utils import delete_model_cache
 
 CANONICAL_COLUMN_VALUES = {
     "prithvida/Splade_PP_en_v1": {
@@ -44,7 +49,8 @@ CANONICAL_COLUMN_VALUES = {
 docs = ["Hello World"]
 
 
-def test_batch_embedding():
+def test_batch_embedding() -> None:
+    is_ci = os.getenv("CI")
     docs_to_embed = docs * 10
 
     for model_name, expected_result in CANONICAL_COLUMN_VALUES.items():
@@ -54,9 +60,12 @@ def test_batch_embedding():
 
         for i, value in enumerate(result.values):
             assert pytest.approx(value, abs=0.001) == expected_result["values"][i]
+        if is_ci:
+            delete_model_cache(model.model._model_dir)
 
 
-def test_single_embedding():
+def test_single_embedding() -> None:
+    is_ci = os.getenv("CI")
     for model_name, expected_result in CANONICAL_COLUMN_VALUES.items():
         model = SparseTextEmbedding(model_name=model_name)
 
@@ -67,11 +76,12 @@ def test_single_embedding():
 
             for i, value in enumerate(result.values):
                 assert pytest.approx(value, abs=0.001) == expected_result["values"][i]
+        if is_ci:
+            delete_model_cache(model.model._model_dir)
 
 
-def test_parallel_processing():
-    import numpy as np
-
+def test_parallel_processing() -> None:
+    is_ci = os.getenv("CI")
     model = SparseTextEmbedding(model_name="prithivida/Splade_PP_en_v1")
     docs = ["hello world", "flag embedding"] * 30
     sparse_embeddings_duo = list(model.embed(docs, batch_size=10, parallel=2))
@@ -93,9 +103,93 @@ def test_parallel_processing():
             == sparse_embedding_duo.indices.tolist()
             == sparse_embedding_all.indices.tolist()
         )
-        assert np.allclose(
-            sparse_embedding.values, sparse_embedding_duo.values, atol=1e-3
-        )
-        assert np.allclose(
-            sparse_embedding.values, sparse_embedding_all.values, atol=1e-3
-        )
+        assert np.allclose(sparse_embedding.values, sparse_embedding_duo.values, atol=1e-3)
+        assert np.allclose(sparse_embedding.values, sparse_embedding_all.values, atol=1e-3)
+
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
+
+
+@pytest.fixture
+def bm25_instance() -> None:
+    ci = os.getenv("CI", True)
+    model = Bm25("Qdrant/bm25", language="english")
+    yield model
+    if ci:
+        delete_model_cache(model._model_dir)
+
+
+def test_stem_with_stopwords_and_punctuation(bm25_instance: Bm25) -> None:
+    # Setup
+    bm25_instance.stopwords = {"the", "is", "a"}
+    bm25_instance.punctuation = {".", ",", "!"}
+
+    # Test data
+    tokens = ["The", "quick", "brown", "fox", "is", "a", "test", "sentence", ".", "!"]
+
+    # Execute
+    result = bm25_instance._stem(tokens)
+
+    # Assert
+    expected = ["quick", "brown", "fox", "test", "sentenc"]
+    assert result == expected, f"Expected {expected}, but got {result}"
+
+
+def test_stem_case_insensitive_stopwords(bm25_instance: Bm25) -> None:
+    # Setup
+    bm25_instance.stopwords = {"the", "is", "a"}
+    bm25_instance.punctuation = {".", ",", "!"}
+
+    # Test data
+    tokens = ["THE", "Quick", "Brown", "Fox", "IS", "A", "Test", "Sentence", ".", "!"]
+
+    # Execute
+    result = bm25_instance._stem(tokens)
+
+    # Assert
+    expected = ["quick", "brown", "fox", "test", "sentenc"]
+    assert result == expected, f"Expected {expected}, but got {result}"
+
+
+@pytest.mark.parametrize("disable_stemmer", [True, False])
+def test_disable_stemmer_behavior(disable_stemmer: bool) -> None:
+    # Setup
+    model = Bm25("Qdrant/bm25", language="english", disable_stemmer=disable_stemmer)
+    model.stopwords = {"the", "is", "a"}
+    model.punctuation = {".", ",", "!"}
+
+    # Test data
+    tokens = ["The", "quick", "brown", "fox", "is", "a", "test", "sentence", ".", "!"]
+
+    # Execute
+    result = model._stem(tokens)
+
+    # Assert
+    if disable_stemmer:
+        expected = ["quick", "brown", "fox", "test", "sentence"]  # no stemming, lower case only
+    else:
+        expected = ["quick", "brown", "fox", "test", "sentenc"]
+    assert result == expected, f"Expected {expected}, but got {result}"
+
+
+@pytest.mark.parametrize(
+    "model_name",
+    ["prithivida/Splade_PP_en_v1"],
+)
+def test_lazy_load(model_name: str) -> None:
+    is_ci = os.getenv("CI")
+    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))
+
+    if is_ci:
+        delete_model_cache(model.model._model_dir)

tests/test_text_cross_encoder.py

@@ -0,0 +1,119 @@
+import os
+
+import numpy as np
+import pytest
+
+from fastembed.rerank.cross_encoder import TextCrossEncoder
+from tests.utils import delete_model_cache
+
+CANONICAL_SCORE_VALUES = {
+    "Xenova/ms-marco-MiniLM-L-6-v2": np.array([8.500708, -2.541011]),
+    "Xenova/ms-marco-MiniLM-L-12-v2": np.array([9.330912, -2.0380247]),
+    "BAAI/bge-reranker-base": np.array([6.15733337, -3.65939403]),
+    "jinaai/jina-reranker-v1-tiny-en": np.array([2.5911, 0.1122]),
+    "jinaai/jina-reranker-v1-turbo-en": np.array([1.8295, -2.8908]),
+    "jinaai/jina-reranker-v2-base-multilingual": np.array([1.6533, -1.6455]),
+}
+
+SELECTED_MODELS = {
+    "Xenova": "Xenova/ms-marco-MiniLM-L-6-v2",
+    "BAAI": "BAAI/bge-reranker-base",
+    "jinaai": "jinaai/jina-reranker-v1-tiny-en",
+}
+
+
+@pytest.mark.parametrize(
+    "model_name",
+    [model_name for model_name in CANONICAL_SCORE_VALUES],
+)
+def test_rerank(model_name: str) -> None:
+    is_ci = os.getenv("CI")
+
+    model = TextCrossEncoder(model_name=model_name)
+
+    query = "What is the capital of France?"
+    documents = ["Paris is the capital of France.", "Berlin is the capital of Germany."]
+    scores = np.array(list(model.rerank(query, documents)))
+
+    pairs = [(query, doc) for doc in documents]
+    scores2 = np.array(list(model.rerank_pairs(pairs)))
+    assert np.allclose(
+        scores, scores2, atol=1e-5
+    ), f"Model: {model_name}, Scores: {scores}, Scores2: {scores2}"
+
+    canonical_scores = CANONICAL_SCORE_VALUES[model_name]
+    assert np.allclose(
+        scores, canonical_scores, atol=1e-3
+    ), f"Model: {model_name}, Scores: {scores}, Expected: {canonical_scores}"
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
+
+
+@pytest.mark.parametrize(
+    "model_name",
+    [model_name for model_name in SELECTED_MODELS.values()],
+)
+def test_batch_rerank(model_name: str) -> None:
+    is_ci = os.getenv("CI")
+
+    model = TextCrossEncoder(model_name=model_name)
+
+    query = "What is the capital of France?"
+    documents = ["Paris is the capital of France.", "Berlin is the capital of Germany."] * 50
+    scores = np.array(list(model.rerank(query, documents, batch_size=10)))
+
+    pairs = [(query, doc) for doc in documents]
+    scores2 = np.array(list(model.rerank_pairs(pairs)))
+    assert np.allclose(
+        scores, scores2, atol=1e-5
+    ), f"Model: {model_name}, Scores: {scores}, Scores2: {scores2}"
+
+    canonical_scores = np.tile(CANONICAL_SCORE_VALUES[model_name], 50)
+
+    assert scores.shape == canonical_scores.shape, f"Unexpected shape for model {model_name}"
+    assert np.allclose(
+        scores, canonical_scores, atol=1e-3
+    ), f"Model: {model_name}, Scores: {scores}, Expected: {canonical_scores}"
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
+
+
+@pytest.mark.parametrize(
+    "model_name",
+    ["Xenova/ms-marco-MiniLM-L-6-v2"],
+)
+def test_lazy_load(model_name: str) -> None:
+    is_ci = os.getenv("CI")
+    model = TextCrossEncoder(model_name=model_name, lazy_load=True)
+    assert not hasattr(model.model, "model")
+    query = "What is the capital of France?"
+    documents = ["Paris is the capital of France.", "Berlin is the capital of Germany."]
+    list(model.rerank(query, documents))
+    assert hasattr(model.model, "model")
+
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
+
+
+@pytest.mark.parametrize(
+    "model_name",
+    [model_name for model_name in SELECTED_MODELS.values()],
+)
+def test_rerank_pairs_parallel(model_name: str) -> None:
+    is_ci = os.getenv("CI")
+
+    model = TextCrossEncoder(model_name=model_name)
+    query = "What is the capital of France?"
+    documents = ["Paris is the capital of France.", "Berlin is the capital of Germany."] * 10
+    pairs = [(query, doc) for doc in documents]
+    scores_parallel = np.array(list(model.rerank_pairs(pairs, parallel=2, batch_size=10)))
+    scores_sequential = np.array(list(model.rerank_pairs(pairs, batch_size=10)))
+    assert np.allclose(
+        scores_parallel, scores_sequential, atol=1e-5
+    ), f"Model: {model_name}, Scores (Parallel): {scores_parallel}, Scores (Sequential): {scores_sequential}"
+    canonical_scores = CANONICAL_SCORE_VALUES[model_name]
+    assert np.allclose(
+        scores_parallel[: len(canonical_scores)], canonical_scores, atol=1e-3
+    ), f"Model: {model_name}, Scores (Parallel): {scores_parallel}, Expected: {canonical_scores}"
+    if is_ci:
+        delete_model_cache(model.model._model_dir)

tests/test_text_multitask_embeddings.py

@@ -0,0 +1,253 @@
+import os
+
+import numpy as np
+import pytest
+
+from fastembed import TextEmbedding
+from fastembed.text.multitask_embedding import Task
+from tests.utils import delete_model_cache
+
+
+CANONICAL_VECTOR_VALUES = {
+    "jinaai/jina-embeddings-v3": [
+        {
+            "task_id": Task.RETRIEVAL_QUERY,
+            "vectors": np.array(
+                [
+                    [0.0623, -0.0402, 0.1706, -0.0143, 0.0617],
+                    [-0.1064, -0.0733, 0.0353, 0.0096, 0.0667],
+                ]
+            ),
+        },
+        {
+            "task_id": Task.RETRIEVAL_PASSAGE,
+            "vectors": np.array(
+                [
+                    [0.0513, -0.0247, 0.1751, -0.0075, 0.0679],
+                    [-0.0987, -0.0786, 0.09, 0.0087, 0.0577],
+                ]
+            ),
+        },
+        {
+            "task_id": Task.SEPARATION,
+            "vectors": np.array(
+                [
+                    [0.094, -0.1065, 0.1305, 0.0547, 0.0556],
+                    [0.0315, -0.1468, 0.065, 0.0568, 0.0546],
+                ]
+            ),
+        },
+        {
+            "task_id": Task.CLASSIFICATION,
+            "vectors": np.array(
+                [
+                    [0.0606, -0.0877, 0.1384, 0.0065, 0.0722],
+                    [-0.0502, -0.119, 0.032, 0.0514, 0.0689],
+                ]
+            ),
+        },
+        {
+            "task_id": Task.TEXT_MATCHING,
+            "vectors": np.array(
+                [
+                    [0.0911, -0.0341, 0.1305, -0.026, 0.0576],
+                    [-0.1432, -0.05, 0.0133, 0.0464, 0.0789],
+                ]
+            ),
+        },
+    ]
+}
+docs = ["Hello World", "Follow the white rabbit."]
+
+
+def test_batch_embedding():
+    is_ci = os.getenv("CI")
+    docs_to_embed = docs * 10
+    default_task = Task.RETRIEVAL_PASSAGE
+
+    for model_desc in TextEmbedding.list_supported_models():
+        if not is_ci and model_desc["size_in_GB"] > 1:
+            continue
+
+        model_name = model_desc["model"]
+        dim = model_desc["dim"]
+
+        if model_name not in CANONICAL_VECTOR_VALUES.keys():
+            continue
+
+        model = TextEmbedding(model_name=model_name)
+
+        print(f"evaluating {model_name} default task")
+
+        embeddings = list(model.embed(documents=docs_to_embed, batch_size=6))
+        embeddings = np.stack(embeddings, axis=0)
+
+        assert embeddings.shape == (len(docs_to_embed), dim)
+
+        canonical_vector = CANONICAL_VECTOR_VALUES[model_name][default_task]["vectors"]
+        assert np.allclose(
+            embeddings[: len(docs), : canonical_vector.shape[1]], canonical_vector, atol=1e-4
+        ), model_desc["model"]
+
+        if is_ci:
+            delete_model_cache(model.model._model_dir)
+
+
+def test_single_embedding():
+    is_ci = os.getenv("CI")
+
+    for model_desc in TextEmbedding.list_supported_models():
+        if not is_ci and model_desc["size_in_GB"] > 1:
+            continue
+
+        model_name = model_desc["model"]
+        dim = model_desc["dim"]
+
+        if model_name not in CANONICAL_VECTOR_VALUES.keys():
+            continue
+
+        model = TextEmbedding(model_name=model_name)
+
+        for task in CANONICAL_VECTOR_VALUES[model_name]:
+            print(f"evaluating {model_name} task_id: {task['task_id']}")
+
+            embeddings = list(model.embed(documents=docs, task_id=task["task_id"]))
+            embeddings = np.stack(embeddings, axis=0)
+
+            assert embeddings.shape == (len(docs), dim)
+
+            canonical_vector = task["vectors"]
+            assert np.allclose(
+                embeddings[: len(docs), : canonical_vector.shape[1]], canonical_vector, atol=1e-4
+            ), model_desc["model"]
+
+        if is_ci:
+            delete_model_cache(model.model._model_dir)
+
+
+def test_single_embedding_query():
+    is_ci = os.getenv("CI")
+    task_id = Task.RETRIEVAL_QUERY
+
+    for model_desc in TextEmbedding.list_supported_models():
+        if not is_ci and model_desc["size_in_GB"] > 1:
+            continue
+
+        model_name = model_desc["model"]
+        dim = model_desc["dim"]
+
+        if model_name not in CANONICAL_VECTOR_VALUES.keys():
+            continue
+
+        model = TextEmbedding(model_name=model_name)
+
+        print(f"evaluating {model_name} query_embed task_id: {task_id}")
+
+        embeddings = list(model.query_embed(query=docs))
+        embeddings = np.stack(embeddings, axis=0)
+
+        assert embeddings.shape == (len(docs), dim)
+
+        canonical_vector = CANONICAL_VECTOR_VALUES[model_name][task_id]["vectors"]
+        assert np.allclose(
+            embeddings[: len(docs), : canonical_vector.shape[1]], canonical_vector, atol=1e-4
+        ), model_desc["model"]
+
+        if is_ci:
+            delete_model_cache(model.model._model_dir)
+
+
+def test_single_embedding_passage():
+    is_ci = os.getenv("CI")
+    task_id = Task.RETRIEVAL_PASSAGE
+
+    for model_desc in TextEmbedding.list_supported_models():
+        if not is_ci and model_desc["size_in_GB"] > 1:
+            continue
+
+        model_name = model_desc["model"]
+        dim = model_desc["dim"]
+
+        if model_name not in CANONICAL_VECTOR_VALUES.keys():
+            continue
+
+        model = TextEmbedding(model_name=model_name)
+
+        print(f"evaluating {model_name} passage_embed task_id: {task_id}")
+
+        embeddings = list(model.passage_embed(texts=docs))
+        embeddings = np.stack(embeddings, axis=0)
+
+        assert embeddings.shape == (len(docs), dim)
+
+        canonical_vector = CANONICAL_VECTOR_VALUES[model_name][task_id]["vectors"]
+        assert np.allclose(
+            embeddings[: len(docs), : canonical_vector.shape[1]], canonical_vector, atol=1e-4
+        ), model_desc["model"]
+
+        if is_ci:
+            delete_model_cache(model.model._model_dir)
+
+
+def test_parallel_processing():
+    is_ci = os.getenv("CI")
+
+    docs = ["Hello World", "Follow the white rabbit."] * 10
+
+    model_name = "jinaai/jina-embeddings-v3"
+    dim = 1024
+
+    model = TextEmbedding(model_name=model_name)
+
+    task_id = Task.SEPARATION
+    embeddings_1 = list(model.embed(docs, batch_size=10, parallel=None, task_id=task_id))
+    embeddings_1 = np.stack(embeddings_1, axis=0)
+
+    embeddings_2 = list(model.embed(docs, batch_size=10, parallel=1, task_id=task_id))
+    embeddings_2 = np.stack(embeddings_2, axis=0)
+
+    assert embeddings_1.shape[0] == len(docs) and embeddings_1.shape[-1] == dim
+    assert np.allclose(embeddings_1, embeddings_2, atol=1e-4)
+
+    canonical_vector = CANONICAL_VECTOR_VALUES[model_name][task_id]["vectors"]
+    assert np.allclose(embeddings_2[:2, : canonical_vector.shape[1]], canonical_vector, atol=1e-4)
+
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
+
+
+def test_task_assignment():
+    is_ci = os.getenv("CI")
+
+    for model_desc in TextEmbedding.list_supported_models():
+        if not is_ci and model_desc["size_in_GB"] > 1:
+            continue
+
+        model_name = model_desc["model"]
+        if model_name not in CANONICAL_VECTOR_VALUES.keys():
+            continue
+
+        model = TextEmbedding(model_name=model_name)
+
+        for i, task_id in enumerate(Task):
+            _ = list(model.embed(documents=docs, batch_size=1, task_id=i))
+            assert model.model._current_task_id == task_id
+
+        if is_ci:
+            delete_model_cache(model.model._model_dir)
+
+
+@pytest.mark.parametrize(
+    "model_name",
+    ["jinaai/jina-embeddings-v3"],
+)
+def test_lazy_load(model_name: str):
+    is_ci = os.getenv("CI")
+    model = TextEmbedding(model_name=model_name, lazy_load=True)
+    assert not hasattr(model.model, "model")
+
+    list(model.embed(docs))
+    assert hasattr(model.model, "model")
+
+    if is_ci:
+        delete_model_cache(model.model._model_dir)

tests/test_text_onnx_embeddings.py

@@ -1,9 +1,11 @@
 import os
+import platform
 
 import numpy as np
 import pytest
 
 from fastembed.text.text_embedding import TextEmbedding
+from tests.utils import delete_model_cache
 
 CANONICAL_VECTOR_VALUES = {
     "BAAI/bge-small-en": np.array([-0.0232, -0.0255, 0.0174, -0.0639, -0.0006]),
@@ -30,22 +32,24 @@ CANONICAL_VECTOR_VALUES = {
         [-0.034478, 0.03102, 0.00673, 0.02611, -0.039362]
     ),
     "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2": np.array(
-        [0.0094, 0.0184, 0.0328, 0.0072, -0.0351]
+        [0.0361, 0.1862, 0.2776, 0.2461, -0.1904]
     ),
-    "intfloat/multilingual-e5-large": np.array([0.0098, 0.0045, 0.0066, -0.0354, 0.0070]),
+    "intfloat/multilingual-e5-large": np.array([0.4544, -0.0968, 0.1054, -1.3753, 0.1500]),
     "sentence-transformers/paraphrase-multilingual-mpnet-base-v2": np.array(
-        [-0.01341097, 0.0416553, -0.00480805, 0.02844842, 0.0505299]
+        [0.0047, 0.1334, -0.0102, 0.0714, 0.1930]
     ),
     "jinaai/jina-embeddings-v2-small-en": np.array([-0.0455, -0.0428, -0.0122, 0.0613, 0.0015]),
     "jinaai/jina-embeddings-v2-base-en": np.array([-0.0332, -0.0509, 0.0287, -0.0043, -0.0077]),
     "jinaai/jina-embeddings-v2-base-de": np.array([-0.0085, 0.0417, 0.0342, 0.0309, -0.0149]),
     "jinaai/jina-embeddings-v2-base-code": np.array([0.0145, -0.0164, 0.0136, -0.0170, 0.0734]),
+    "jinaai/jina-embeddings-v2-base-zh": np.array([0.0381, 0.0286, -0.0231, 0.0052, -0.0151]),
+    "jinaai/jina-embeddings-v2-base-es": np.array([-0.0108, -0.0092, -0.0373, 0.0171, -0.0301]),
     "nomic-ai/nomic-embed-text-v1": np.array([0.3708, 0.2031, -0.3406, -0.2114, -0.3230]),
     "nomic-ai/nomic-embed-text-v1.5": np.array(
         [-0.15407836, -0.03053198, -3.9138033, 0.1910364, 0.13224715]
     ),
     "nomic-ai/nomic-embed-text-v1.5-Q": np.array(
-        [-0.12525563, 0.38030425, -3.961622, 0.04176439, -0.0758301]
+        [0.0802303, 0.3700881, -4.3053818, 0.4431803, -0.271572]
     ),
     "thenlper/gte-large": np.array(
         [-0.01920587, 0.00113156, -0.00708992, -0.00632304, -0.04025577]
@@ -61,14 +65,23 @@ CANONICAL_VECTOR_VALUES = {
     ),
     "snowflake/snowflake-arctic-embed-l": np.array([0.0189, -0.0673, 0.0183, 0.0124, 0.0146]),
     "Qdrant/clip-ViT-B-32-text": np.array([0.0083, 0.0103, -0.0138, 0.0199, -0.0069]),
+    "thenlper/gte-base": np.array([0.0038, 0.0355, 0.0181, 0.0092, 0.0654]),
+    "jinaai/jina-clip-v1": np.array([-0.0862, -0.0101, -0.0056, 0.0375, -0.0472]),
 }
 
+MULTI_TASK_MODELS = ["jinaai/jina-embeddings-v3"]
 
-def test_embedding():
+
+def test_embedding() -> None:
     is_ci = os.getenv("CI")
+    is_mac = platform.system() == "Darwin"
 
     for model_desc in TextEmbedding.list_supported_models():
-        if not is_ci and model_desc["size_in_GB"] > 1:
+        if (
+            (not is_ci and model_desc["size_in_GB"] > 1)
+            or model_desc["model"] in MULTI_TASK_MODELS
+            or (is_mac and model_desc["model"] == "nomic-ai/nomic-embed-text-v1.5-Q")
+        ):
             continue
 
         dim = model_desc["dim"]
@@ -83,13 +96,16 @@ def test_embedding():
         assert np.allclose(
             embeddings[0, : canonical_vector.shape[0]], canonical_vector, atol=1e-3
         ), model_desc["model"]
+        if is_ci:
+            delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize(
     "n_dims,model_name",
     [(384, "BAAI/bge-small-en-v1.5"), (768, "jinaai/jina-embeddings-v2-base-en")],
 )
-def test_batch_embedding(n_dims, model_name):
+def test_batch_embedding(n_dims: int, model_name: str) -> None:
+    is_ci = os.getenv("CI")
     model = TextEmbedding(model_name=model_name)
 
     docs = ["hello world", "flag embedding"] * 100
@@ -97,13 +113,16 @@ def test_batch_embedding(n_dims, model_name):
     embeddings = np.stack(embeddings, axis=0)
 
     assert embeddings.shape == (200, n_dims)
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize(
     "n_dims,model_name",
     [(384, "BAAI/bge-small-en-v1.5"), (768, "jinaai/jina-embeddings-v2-base-en")],
 )
-def test_parallel_processing(n_dims, model_name):
+def test_parallel_processing(n_dims: int, model_name: str) -> None:
+    is_ci = os.getenv("CI")
     model = TextEmbedding(model_name=model_name)
 
     docs = ["hello world", "flag embedding"] * 100
@@ -119,3 +138,28 @@ def test_parallel_processing(n_dims, model_name):
     assert embeddings.shape == (200, n_dims)
     assert np.allclose(embeddings, embeddings_2, atol=1e-3)
     assert np.allclose(embeddings, embeddings_3, atol=1e-3)
+
+    if is_ci:
+        delete_model_cache(model.model._model_dir)
+
+
+@pytest.mark.parametrize(
+    "model_name",
+    ["BAAI/bge-small-en-v1.5"],
+)
+def test_lazy_load(model_name: str) -> None:
+    is_ci = os.getenv("CI")
+    model = TextEmbedding(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 = TextEmbedding(model_name=model_name, lazy_load=True)
+    list(model.query_embed(docs))
+
+    model = TextEmbedding(model_name=model_name, lazy_load=True)
+    list(model.passage_embed(docs))
+
+    if is_ci:
+        delete_model_cache(model.model._model_dir)

tests/utils.py

@@ -0,0 +1,37 @@
+import shutil
+import traceback
+
+from pathlib import Path
+from types import TracebackType
+from typing import Union, Callable, Any, Type
+
+
+def delete_model_cache(model_dir: Union[str, Path]) -> None:
+    """Delete the model cache directory.
+
+    If a model was downloaded from the HuggingFace model hub, then _model_dir is the dir to snapshots, removing
+    it won't help to release the memory, because data is in blobs directory.
+    If a model was downloaded from GCS, then we can just remove model_dir
+
+    Args:
+        model_dir (Union[str, Path]): The path to the model cache directory.
+    """
+
+    def on_error(
+        func: Callable[..., Any],
+        path: str,
+        exc_info: tuple[Type[BaseException], BaseException, TracebackType],
+    ) -> None:
+        print("Failed to remove: ", path)
+        print("Exception: ", exc_info)
+        traceback.print_exception(*exc_info)
+
+    if isinstance(model_dir, str):
+        model_dir = Path(model_dir)
+
+    if model_dir.parent.parent.name.startswith("models--"):
+        model_dir = model_dir.parent.parent
+
+    if model_dir.exists():
+        # todo: PermissionDenied is raised on blobs removal in Windows, with blobs > 2GB
+        shutil.rmtree(model_dir, onerror=on_error)