Bump version

* HF sources for all models

.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: v0.4.2
     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

@@ -14,11 +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
@@ -42,4 +42,4 @@ jobs:
 
       - name: Run pytest
         run: |
-          poetry run pytest
+          poetry run pytest 

NOTICE

@@ -5,6 +5,8 @@ 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
 
 These models are developed by Jina (https://jina.ai/) and are subject to Jina AI's 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
 

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/FastEmbed_GPU.ipynb

@@ -388,8 +388,6 @@
     }
    ],
    "source": [
-    "from typing import List\n",
-    "\n",
     "import numpy as np\n",
     "\n",
     "from fastembed import TextEmbedding\n",
@@ -407,9 +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",
@@ -155,7 +156,7 @@
     "        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]):\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"
@@ -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/examples/Supported_Models.ipynb

@@ -2,38 +2,36 @@
  "cells": [
   {
    "cell_type": "code",
-   "execution_count": 1,
    "metadata": {
     "ExecuteTime": {
-     "end_time": "2024-05-31T18:13:23.806907Z",
-     "start_time": "2024-05-31T18:13:23.797078Z"
+     "end_time": "2024-11-13T09:01:03.324551Z",
+     "start_time": "2024-11-13T09:01:03.234711Z"
     }
    },
-   "outputs": [],
    "source": [
     "%load_ext autoreload\n",
     "%autoreload 2"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 2,
-   "metadata": {
-    "ExecuteTime": {
-     "end_time": "2024-05-31T18:14:31.147674Z",
-     "start_time": "2024-05-31T18:14:31.134015Z"
-    }
-   },
+   ],
    "outputs": [
     {
-     "name": "stderr",
+     "name": "stdout",
      "output_type": "stream",
      "text": [
-      "/home/hossam/.pyenv/versions/.venv/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
-      "  from .autonotebook import tqdm as notebook_tqdm\n"
+      "The autoreload extension is already loaded. To reload it, use:\n",
+      "  %reload_ext autoreload\n"
      ]
     }
    ],
+   "execution_count": 10
+  },
+  {
+   "cell_type": "code",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-11-13T09:01:04.505772Z",
+     "start_time": "2024-11-13T09:01:04.493296Z"
+    }
+   },
    "source": [
     "import pandas as pd\n",
     "\n",
@@ -42,8 +40,11 @@
     "    TextEmbedding,\n",
     "    LateInteractionTextEmbedding,\n",
     "    ImageEmbedding,\n",
-    ")"
-   ]
+    ")\n",
+    "from fastembed.rerank.cross_encoder import TextCrossEncoder"
+   ],
+   "outputs": [],
+   "execution_count": 11
   },
   {
    "cell_type": "markdown",
@@ -54,16 +55,79 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
    "metadata": {
     "ExecuteTime": {
-     "end_time": "2024-05-31T18:13:25.863008Z",
-     "start_time": "2024-05-31T18:13:25.837795Z"
+     "end_time": "2024-11-13T09:01:05.812271Z",
+     "start_time": "2024-11-13T09:01:05.795846Z"
     }
    },
+   "source": [
+    "supported_models = (\n",
+    "    pd.DataFrame(TextEmbedding.list_supported_models())\n",
+    "    .sort_values(\"size_in_GB\")\n",
+    "    .drop(columns=[\"sources\", \"model_file\", \"additional_files\"])\n",
+    "    .reset_index(drop=True)\n",
+    ")\n",
+    "supported_models"
+   ],
    "outputs": [
     {
      "data": {
+      "text/plain": [
+       "                                                model   dim  \\\n",
+       "0                              BAAI/bge-small-en-v1.5   384   \n",
+       "1                              BAAI/bge-small-zh-v1.5   512   \n",
+       "2                 snowflake/snowflake-arctic-embed-xs   384   \n",
+       "3              sentence-transformers/all-MiniLM-L6-v2   384   \n",
+       "4                  jinaai/jina-embeddings-v2-small-en   512   \n",
+       "5                                   BAAI/bge-small-en   384   \n",
+       "6                  snowflake/snowflake-arctic-embed-s   384   \n",
+       "7                    nomic-ai/nomic-embed-text-v1.5-Q   768   \n",
+       "8                               BAAI/bge-base-en-v1.5   768   \n",
+       "9   sentence-transformers/paraphrase-multilingual-...   384   \n",
+       "10                          Qdrant/clip-ViT-B-32-text   512   \n",
+       "11                  jinaai/jina-embeddings-v2-base-de   768   \n",
+       "12                                   BAAI/bge-base-en   768   \n",
+       "13                 snowflake/snowflake-arctic-embed-m   768   \n",
+       "14                     nomic-ai/nomic-embed-text-v1.5   768   \n",
+       "15                  jinaai/jina-embeddings-v2-base-en   768   \n",
+       "16                       nomic-ai/nomic-embed-text-v1   768   \n",
+       "17            snowflake/snowflake-arctic-embed-m-long   768   \n",
+       "18                 mixedbread-ai/mxbai-embed-large-v1  1024   \n",
+       "19                jinaai/jina-embeddings-v2-base-code   768   \n",
+       "20  sentence-transformers/paraphrase-multilingual-...   768   \n",
+       "21                 snowflake/snowflake-arctic-embed-l  1024   \n",
+       "22                                 thenlper/gte-large  1024   \n",
+       "23                             BAAI/bge-large-en-v1.5  1024   \n",
+       "24                     intfloat/multilingual-e5-large  1024   \n",
+       "\n",
+       "                                          description     license  size_in_GB  \n",
+       "0   Text embeddings, Unimodal (text), English, 512...         mit       0.067  \n",
+       "1   Text embeddings, Unimodal (text), Chinese, 512...         mit       0.090  \n",
+       "2   Text embeddings, Unimodal (text), English, 512...  apache-2.0       0.090  \n",
+       "3   Text embeddings, Unimodal (text), English, 256...  apache-2.0       0.090  \n",
+       "4   Text embeddings, Unimodal (text), English, 819...  apache-2.0       0.120  \n",
+       "5   Text embeddings, Unimodal (text), English, 512...         mit       0.130  \n",
+       "6   Text embeddings, Unimodal (text), English, 512...  apache-2.0       0.130  \n",
+       "7   Text embeddings, Multimodal (text, image), Eng...  apache-2.0       0.130  \n",
+       "8   Text embeddings, Unimodal (text), English, 512...         mit       0.210  \n",
+       "9   Text embeddings, Unimodal (text), Multilingual...  apache-2.0       0.220  \n",
+       "10  Text embeddings, Multimodal (text&image), Engl...         mit       0.250  \n",
+       "11  Text embeddings, Unimodal (text), Multilingual...  apache-2.0       0.320  \n",
+       "12  Text embeddings, Unimodal (text), English, 512...         mit       0.420  \n",
+       "13  Text embeddings, Unimodal (text), English, 512...  apache-2.0       0.430  \n",
+       "14  Text embeddings, Multimodal (text, image), Eng...  apache-2.0       0.520  \n",
+       "15  Text embeddings, Unimodal (text), English, 819...  apache-2.0       0.520  \n",
+       "16  Text embeddings, Multimodal (text, image), Eng...  apache-2.0       0.520  \n",
+       "17  Text embeddings, Unimodal (text), English, 204...  apache-2.0       0.540  \n",
+       "18  Text embeddings, Unimodal (text), English, 512...  apache-2.0       0.640  \n",
+       "19  Text embeddings, Unimodal (text), Multilingual...  apache-2.0       0.640  \n",
+       "20  Text embeddings, Unimodal (text), Multilingual...  apache-2.0       1.000  \n",
+       "21  Text embeddings, Unimodal (text), English, 512...  apache-2.0       1.020  \n",
+       "22  Text embeddings, Unimodal (text), English, 512...         mit       1.200  \n",
+       "23  Text embeddings, Unimodal (text), English, 512...         mit       1.200  \n",
+       "24  Text embeddings, Unimodal (text), Multilingual...         mit       2.240  "
+      ],
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
@@ -294,77 +358,14 @@
        "  </tbody>\n",
        "</table>\n",
        "</div>"
-      ],
-      "text/plain": [
-       "                                                model   dim  \\\n",
-       "0                              BAAI/bge-small-en-v1.5   384   \n",
-       "1                              BAAI/bge-small-zh-v1.5   512   \n",
-       "2                 snowflake/snowflake-arctic-embed-xs   384   \n",
-       "3              sentence-transformers/all-MiniLM-L6-v2   384   \n",
-       "4                  jinaai/jina-embeddings-v2-small-en   512   \n",
-       "5                                   BAAI/bge-small-en   384   \n",
-       "6                  snowflake/snowflake-arctic-embed-s   384   \n",
-       "7                    nomic-ai/nomic-embed-text-v1.5-Q   768   \n",
-       "8                               BAAI/bge-base-en-v1.5   768   \n",
-       "9   sentence-transformers/paraphrase-multilingual-...   384   \n",
-       "10                          Qdrant/clip-ViT-B-32-text   512   \n",
-       "11                  jinaai/jina-embeddings-v2-base-de   768   \n",
-       "12                                   BAAI/bge-base-en   768   \n",
-       "13                 snowflake/snowflake-arctic-embed-m   768   \n",
-       "14                     nomic-ai/nomic-embed-text-v1.5   768   \n",
-       "15                  jinaai/jina-embeddings-v2-base-en   768   \n",
-       "16                       nomic-ai/nomic-embed-text-v1   768   \n",
-       "17            snowflake/snowflake-arctic-embed-m-long   768   \n",
-       "18                 mixedbread-ai/mxbai-embed-large-v1  1024   \n",
-       "19                jinaai/jina-embeddings-v2-base-code   768   \n",
-       "20  sentence-transformers/paraphrase-multilingual-...   768   \n",
-       "21                 snowflake/snowflake-arctic-embed-l  1024   \n",
-       "22                                 thenlper/gte-large  1024   \n",
-       "23                             BAAI/bge-large-en-v1.5  1024   \n",
-       "24                     intfloat/multilingual-e5-large  1024   \n",
-       "\n",
-       "                                          description     license  size_in_GB  \n",
-       "0   Text embeddings, Unimodal (text), English, 512...         mit       0.067  \n",
-       "1   Text embeddings, Unimodal (text), Chinese, 512...         mit       0.090  \n",
-       "2   Text embeddings, Unimodal (text), English, 512...  apache-2.0       0.090  \n",
-       "3   Text embeddings, Unimodal (text), English, 256...  apache-2.0       0.090  \n",
-       "4   Text embeddings, Unimodal (text), English, 819...  apache-2.0       0.120  \n",
-       "5   Text embeddings, Unimodal (text), English, 512...         mit       0.130  \n",
-       "6   Text embeddings, Unimodal (text), English, 512...  apache-2.0       0.130  \n",
-       "7   Text embeddings, Multimodal (text, image), Eng...  apache-2.0       0.130  \n",
-       "8   Text embeddings, Unimodal (text), English, 512...         mit       0.210  \n",
-       "9   Text embeddings, Unimodal (text), Multilingual...  apache-2.0       0.220  \n",
-       "10  Text embeddings, Multimodal (text&image), Engl...         mit       0.250  \n",
-       "11  Text embeddings, Unimodal (text), Multilingual...  apache-2.0       0.320  \n",
-       "12  Text embeddings, Unimodal (text), English, 512...         mit       0.420  \n",
-       "13  Text embeddings, Unimodal (text), English, 512...  apache-2.0       0.430  \n",
-       "14  Text embeddings, Multimodal (text, image), Eng...  apache-2.0       0.520  \n",
-       "15  Text embeddings, Unimodal (text), English, 819...  apache-2.0       0.520  \n",
-       "16  Text embeddings, Multimodal (text, image), Eng...  apache-2.0       0.520  \n",
-       "17  Text embeddings, Unimodal (text), English, 204...  apache-2.0       0.540  \n",
-       "18  Text embeddings, Unimodal (text), English, 512...  apache-2.0       0.640  \n",
-       "19  Text embeddings, Unimodal (text), Multilingual...  apache-2.0       0.640  \n",
-       "20  Text embeddings, Unimodal (text), Multilingual...  apache-2.0       1.000  \n",
-       "21  Text embeddings, Unimodal (text), English, 512...  apache-2.0       1.020  \n",
-       "22  Text embeddings, Unimodal (text), English, 512...         mit       1.200  \n",
-       "23  Text embeddings, Unimodal (text), English, 512...         mit       1.200  \n",
-       "24  Text embeddings, Unimodal (text), Multilingual...         mit       2.240  "
       ]
      },
-     "execution_count": 3,
+     "execution_count": 12,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
-   "source": [
-    "supported_models = (\n",
-    "    pd.DataFrame(TextEmbedding.list_supported_models())\n",
-    "    .sort_values(\"size_in_GB\")\n",
-    "    .drop(columns=[\"sources\", \"model_file\", \"additional_files\"])\n",
-    "    .reset_index(drop=True)\n",
-    ")\n",
-    "supported_models"
-   ]
+   "execution_count": 12
   },
   {
    "cell_type": "markdown",
@@ -375,16 +376,42 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 4,
    "metadata": {
     "ExecuteTime": {
-     "end_time": "2024-05-31T18:13:27.124747Z",
-     "start_time": "2024-05-31T18:13:27.096212Z"
+     "end_time": "2024-11-13T09:01:07.038954Z",
+     "start_time": "2024-11-13T09:01:07.019656Z"
     }
    },
+   "source": [
+    "(\n",
+    "    pd.DataFrame(SparseTextEmbedding.list_supported_models())\n",
+    "    .sort_values(\"size_in_GB\")\n",
+    "    .drop(columns=[\"sources\", \"model_file\", \"additional_files\"])\n",
+    "    .reset_index(drop=True)\n",
+    ")"
+   ],
    "outputs": [
     {
      "data": {
+      "text/plain": [
+       "                                     model  vocab_size  \\\n",
+       "0                              Qdrant/bm25         NaN   \n",
+       "1  Qdrant/bm42-all-minilm-l6-v2-attentions     30522.0   \n",
+       "2               prithivida/Splade_PP_en_v1     30522.0   \n",
+       "3                prithvida/Splade_PP_en_v1     30522.0   \n",
+       "\n",
+       "                                         description     license  size_in_GB  \\\n",
+       "0  BM25 as sparse embeddings meant to be used wit...  apache-2.0       0.010   \n",
+       "1  Light sparse embedding model, which assigns an...  apache-2.0       0.090   \n",
+       "2  Independent Implementation of SPLADE++ Model f...  apache-2.0       0.532   \n",
+       "3  Independent Implementation of SPLADE++ Model f...  apache-2.0       0.532   \n",
+       "\n",
+       "  requires_idf  \n",
+       "0         True  \n",
+       "1         True  \n",
+       "2          NaN  \n",
+       "3          NaN  "
+      ],
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
@@ -452,40 +479,14 @@
        "  </tbody>\n",
        "</table>\n",
        "</div>"
-      ],
-      "text/plain": [
-       "                                     model  vocab_size  \\\n",
-       "0                              Qdrant/bm25         NaN   \n",
-       "1  Qdrant/bm42-all-minilm-l6-v2-attentions     30522.0   \n",
-       "2               prithivida/Splade_PP_en_v1     30522.0   \n",
-       "3                prithvida/Splade_PP_en_v1     30522.0   \n",
-       "\n",
-       "                                         description     license  size_in_GB  \\\n",
-       "0  BM25 as sparse embeddings meant to be used wit...  apache-2.0       0.010   \n",
-       "1  Light sparse embedding model, which assigns an...  apache-2.0       0.090   \n",
-       "2  Independent Implementation of SPLADE++ Model f...  apache-2.0       0.532   \n",
-       "3  Independent Implementation of SPLADE++ Model f...  apache-2.0       0.532   \n",
-       "\n",
-       "  requires_idf  \n",
-       "0         True  \n",
-       "1         True  \n",
-       "2          NaN  \n",
-       "3          NaN  "
       ]
      },
-     "execution_count": 4,
+     "execution_count": 13,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
-   "source": [
-    "(\n",
-    "    pd.DataFrame(SparseTextEmbedding.list_supported_models())\n",
-    "    .sort_values(\"size_in_GB\")\n",
-    "    .drop(columns=[\"sources\", \"model_file\", \"additional_files\"])\n",
-    "    .reset_index(drop=True)\n",
-    ")"
-   ]
+   "execution_count": 13
   },
   {
    "cell_type": "markdown",
@@ -498,17 +499,40 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
    "metadata": {
+    "collapsed": false,
     "ExecuteTime": {
-     "end_time": "2024-05-31T18:14:34.370252Z",
-     "start_time": "2024-05-31T18:14:34.354270Z"
-    },
-    "collapsed": false
+     "end_time": "2024-11-13T09:01:08.074442Z",
+     "start_time": "2024-11-13T09:01:08.056138Z"
+    }
    },
+   "source": [
+    "(\n",
+    "    pd.DataFrame(LateInteractionTextEmbedding.list_supported_models())\n",
+    "    .sort_values(\"size_in_GB\")\n",
+    "    .drop(columns=[\"sources\", \"model_file\"])\n",
+    "    .reset_index(drop=True)\n",
+    ")"
+   ],
    "outputs": [
     {
      "data": {
+      "text/plain": [
+       "                                   model  dim  \\\n",
+       "0  answerdotai/answerai-colbert-small-v1   96   \n",
+       "1                 colbert-ir/colbertv2.0  128   \n",
+       "2                 jinaai/jina-colbert-v2  128   \n",
+       "\n",
+       "                                         description       license  \\\n",
+       "0  Text embeddings, Unimodal (text), Multilingual...    apache-2.0   \n",
+       "1                             Late interaction model           mit   \n",
+       "2  New model that expands capabilities of colbert...  cc-by-nc-4.0   \n",
+       "\n",
+       "   size_in_GB        additional_files  \n",
+       "0        0.13                     NaN  \n",
+       "1        0.44                     NaN  \n",
+       "2        2.24  [onnx/model.onnx_data]  "
+      ],
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
@@ -558,7 +582,7 @@
        "    <tr>\n",
        "      <th>2</th>\n",
        "      <td>jinaai/jina-colbert-v2</td>\n",
-       "      <td>1024</td>\n",
+       "      <td>128</td>\n",
        "      <td>New model that expands capabilities of colbert...</td>\n",
        "      <td>cc-by-nc-4.0</td>\n",
        "      <td>2.24</td>\n",
@@ -567,37 +591,14 @@
        "  </tbody>\n",
        "</table>\n",
        "</div>"
-      ],
-      "text/plain": [
-       "                                   model   dim  \\\n",
-       "0  answerdotai/answerai-colbert-small-v1    96   \n",
-       "1                 colbert-ir/colbertv2.0   128   \n",
-       "2                 jinaai/jina-colbert-v2  1024   \n",
-       "\n",
-       "                                         description       license  \\\n",
-       "0  Text embeddings, Unimodal (text), Multilingual...    apache-2.0   \n",
-       "1                             Late interaction model           mit   \n",
-       "2  New model that expands capabilities of colbert...  cc-by-nc-4.0   \n",
-       "\n",
-       "   size_in_GB        additional_files  \n",
-       "0        0.13                     NaN  \n",
-       "1        0.44                     NaN  \n",
-       "2        2.24  [onnx/model.onnx_data]  "
       ]
      },
-     "execution_count": 5,
+     "execution_count": 14,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
-   "source": [
-    "(\n",
-    "    pd.DataFrame(LateInteractionTextEmbedding.list_supported_models())\n",
-    "    .sort_values(\"size_in_GB\")\n",
-    "    .drop(columns=[\"sources\", \"model_file\"])\n",
-    "    .reset_index(drop=True)\n",
-    ")"
-   ]
+   "execution_count": 14
   },
   {
    "cell_type": "markdown",
@@ -610,17 +611,37 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 6,
    "metadata": {
+    "collapsed": false,
     "ExecuteTime": {
-     "end_time": "2024-05-31T18:14:42.501881Z",
-     "start_time": "2024-05-31T18:14:42.484726Z"
-    },
-    "collapsed": false
+     "end_time": "2024-11-13T09:01:09.171647Z",
+     "start_time": "2024-11-13T09:01:09.150940Z"
+    }
    },
+   "source": [
+    "(\n",
+    "    pd.DataFrame(ImageEmbedding.list_supported_models())\n",
+    "    .sort_values(\"size_in_GB\")\n",
+    "    .drop(columns=[\"sources\", \"model_file\"])\n",
+    "    .reset_index(drop=True)\n",
+    ")"
+   ],
    "outputs": [
     {
      "data": {
+      "text/plain": [
+       "                         model   dim  \\\n",
+       "0         Qdrant/resnet50-onnx  2048   \n",
+       "1  Qdrant/clip-ViT-B-32-vision   512   \n",
+       "2       Qdrant/Unicom-ViT-B-32   512   \n",
+       "3       Qdrant/Unicom-ViT-B-16   768   \n",
+       "\n",
+       "                                         description     license  size_in_GB  \n",
+       "0      Image embeddings, Unimodal (image), 2016 year  apache-2.0        0.10  \n",
+       "1  Image embeddings, Multimodal (text&image), 202...         mit        0.34  \n",
+       "2  Image embeddings, Multimodal (text&image), 202...  apache-2.0        0.48  \n",
+       "3  Image embeddings (more detailed than Unicom-Vi...  apache-2.0        0.82  "
+      ],
       "text/html": [
        "<div>\n",
        "<style scoped>\n",
@@ -683,39 +704,149 @@
        "  </tbody>\n",
        "</table>\n",
        "</div>"
-      ],
-      "text/plain": [
-       "                         model   dim  \\\n",
-       "0         Qdrant/resnet50-onnx  2048   \n",
-       "1  Qdrant/clip-ViT-B-32-vision   512   \n",
-       "2       Qdrant/Unicom-ViT-B-32   512   \n",
-       "3       Qdrant/Unicom-ViT-B-16   768   \n",
-       "\n",
-       "                                         description     license  size_in_GB  \n",
-       "0      Image embeddings, Unimodal (image), 2016 year  apache-2.0        0.10  \n",
-       "1  Image embeddings, Multimodal (text&image), 202...         mit        0.34  \n",
-       "2  Image embeddings, Multimodal (text&image), 202...  apache-2.0        0.48  \n",
-       "3  Image embeddings (more detailed than Unicom-Vi...  apache-2.0        0.82  "
       ]
      },
-     "execution_count": 6,
+     "execution_count": 15,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
+   "execution_count": 15
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Supported Rerank Cross Encoder Models"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-11-13T09:01:10.313943Z",
+     "start_time": "2024-11-13T09:01:10.298428Z"
+    }
+   },
    "source": [
     "(\n",
-    "    pd.DataFrame(ImageEmbedding.list_supported_models())\n",
+    "    pd.DataFrame(TextCrossEncoder.list_supported_models())\n",
     "    .sort_values(\"size_in_GB\")\n",
     "    .drop(columns=[\"sources\", \"model_file\"])\n",
     "    .reset_index(drop=True)\n",
     ")"
-   ]
+   ],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "                                       model  size_in_GB  \\\n",
+       "0              Xenova/ms-marco-MiniLM-L-6-v2        0.08   \n",
+       "1             Xenova/ms-marco-MiniLM-L-12-v2        0.12   \n",
+       "2            jinaai/jina-reranker-v1-tiny-en        0.13   \n",
+       "3           jinaai/jina-reranker-v1-turbo-en        0.15   \n",
+       "4                     BAAI/bge-reranker-base        1.04   \n",
+       "5  jinaai/jina-reranker-v2-base-multilingual        1.11   \n",
+       "\n",
+       "                                         description       license  \n",
+       "0  MiniLM-L-6-v2 model optimized for re-ranking t...    apache-2.0  \n",
+       "1  MiniLM-L-12-v2 model optimized for re-ranking ...    apache-2.0  \n",
+       "2  Designed for blazing-fast re-ranking with 8K c...    apache-2.0  \n",
+       "3  Designed for blazing-fast re-ranking with 8K c...    apache-2.0  \n",
+       "4  BGE reranker base model for cross-encoder re-r...           mit  \n",
+       "5  A multi-lingual reranker model for cross-encod...  cc-by-nc-4.0  "
+      ],
+      "text/html": [
+       "<div>\n",
+       "<style scoped>\n",
+       "    .dataframe tbody tr th:only-of-type {\n",
+       "        vertical-align: middle;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe tbody tr th {\n",
+       "        vertical-align: top;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe thead th {\n",
+       "        text-align: right;\n",
+       "    }\n",
+       "</style>\n",
+       "<table border=\"1\" class=\"dataframe\">\n",
+       "  <thead>\n",
+       "    <tr style=\"text-align: right;\">\n",
+       "      <th></th>\n",
+       "      <th>model</th>\n",
+       "      <th>size_in_GB</th>\n",
+       "      <th>description</th>\n",
+       "      <th>license</th>\n",
+       "    </tr>\n",
+       "  </thead>\n",
+       "  <tbody>\n",
+       "    <tr>\n",
+       "      <th>0</th>\n",
+       "      <td>Xenova/ms-marco-MiniLM-L-6-v2</td>\n",
+       "      <td>0.08</td>\n",
+       "      <td>MiniLM-L-6-v2 model optimized for re-ranking t...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>1</th>\n",
+       "      <td>Xenova/ms-marco-MiniLM-L-12-v2</td>\n",
+       "      <td>0.12</td>\n",
+       "      <td>MiniLM-L-12-v2 model optimized for re-ranking ...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>2</th>\n",
+       "      <td>jinaai/jina-reranker-v1-tiny-en</td>\n",
+       "      <td>0.13</td>\n",
+       "      <td>Designed for blazing-fast re-ranking with 8K c...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>3</th>\n",
+       "      <td>jinaai/jina-reranker-v1-turbo-en</td>\n",
+       "      <td>0.15</td>\n",
+       "      <td>Designed for blazing-fast re-ranking with 8K c...</td>\n",
+       "      <td>apache-2.0</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>4</th>\n",
+       "      <td>BAAI/bge-reranker-base</td>\n",
+       "      <td>1.04</td>\n",
+       "      <td>BGE reranker base model for cross-encoder re-r...</td>\n",
+       "      <td>mit</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>5</th>\n",
+       "      <td>jinaai/jina-reranker-v2-base-multilingual</td>\n",
+       "      <td>1.11</td>\n",
+       "      <td>A multi-lingual reranker model for cross-encod...</td>\n",
+       "      <td>cc-by-nc-4.0</td>\n",
+       "    </tr>\n",
+       "  </tbody>\n",
+       "</table>\n",
+       "</div>"
+      ]
+     },
+     "execution_count": 16,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "execution_count": 16
+  },
+  {
+   "metadata": {},
+   "cell_type": "code",
+   "outputs": [],
+   "execution_count": null,
+   "source": ""
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": ".venv",
+   "display_name": "Python 3.8.18 ('base')",
    "language": "python",
    "name": "python3"
   },
@@ -729,9 +860,14 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.10.15"
+   "version": "3.11.8"
   },
-  "orig_nbformat": 4
+  "orig_nbformat": 4,
+  "vscode": {
+   "interpreter": {
+    "hash": "c4a27af61e455bc18dcf16f5867a2ff0402fa12b01dd0f6ce3a79ae73ad15e91"
+   }
+  }
  },
  "nbformat": 4,
  "nbformat_minor": 2

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

@@ -3,27 +3,31 @@ import time
 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.utils import (
+    RepositoryNotFoundError,
+    disable_progress_bars,
+    enable_progress_bars,
+)
 from loguru import logger
 from tqdm import tqdm
 
 
 class ModelManagement:
     @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 +38,7 @@ 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():
@@ -93,8 +97,8 @@ class ModelManagement:
     def download_files_from_huggingface(
         cls,
         hf_source_repo: str,
-        cache_dir: Optional[str] = None,
-        extra_patterns: Optional[List[str]] = None,
+        cache_dir: str,
+        extra_patterns: Optional[list[str]] = None,
         local_files_only: bool = False,
         **kwargs,
     ) -> str:
@@ -103,7 +107,7 @@ class ModelManagement:
         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 (Optional[list[str]]): extra patterns to allow in the snapshot download, typically
                 includes the required model files.
             local_files_only (bool, optional): Whether to only use local files. Defaults to False.
         Returns:
@@ -119,6 +123,12 @@ class ModelManagement:
         if extra_patterns is not None:
             allow_patterns.extend(extra_patterns)
 
+        snapshot_dir = Path(cache_dir) / f"models--{hf_source_repo.replace('/', '--')}"
+        is_cached = snapshot_dir.exists()
+
+        if is_cached:
+            disable_progress_bars()
+
         return snapshot_download(
             repo_id=hf_source_repo,
             allow_patterns=allow_patterns,
@@ -211,13 +221,13 @@ class ModelManagement:
 
     @classmethod
     def download_model(
-        cls, model: Dict[str, Any], cache_dir: Path, retries: int = 3, **kwargs
+        cls, model: dict[str, Any], cache_dir: Path, retries: int = 3, **kwargs
     ) -> 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:
                 ```
                 {
@@ -265,6 +275,8 @@ class ModelManagement:
                             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(

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
@@ -43,8 +33,8 @@ 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
+    ) -> dict[str, np.ndarray]:
         """
         Preprocess the onnx input.
         """
@@ -62,6 +52,13 @@ class OnnxModel(Generic[T]):
         model_path = model_dir / model_file
         # List of Execution Providers: https://onnxruntime.ai/docs/execution-providers
 
+        if cuda and providers is not None:
+            warnings.warn(
+                f"`cuda` and `providers` are mutually exclusive parameters, cuda: {cuda}, providers: {providers}",
+                category=UserWarning,
+                stacklevel=6,
+            )
+
         if providers is not None:
             onnx_providers = list(providers)
         elif cuda:
@@ -131,5 +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/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
 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
@@ -13,4 +13,4 @@ PathInput: TypeAlias = Union[str, os.PathLike]
 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 +1,13 @@
 import os
-import tempfile
-from itertools import islice
-from pathlib import Path
-from typing import Generator, Iterable, Optional, Union
-import unicodedata
 import sys
-import numpy as np
 import re
-from typing import Set
+import tempfile
+import unicodedata
+from pathlib import Path
+from itertools import islice
+from typing import Generator, Iterable, Optional, Union
+
+import numpy as np
 
 
 def normalize(input_array, p=2, dim=1, eps=1e-12) -> np.ndarray:
@@ -45,7 +45,7 @@ def define_cache_dir(cache_dir: Optional[str] = None) -> Path:
     return cache_path
 
 
-def get_all_punctuation() -> Set[str]:
+def get_all_punctuation() -> set[str]:
     return set(
         chr(i) for i in range(sys.maxunicode) if unicodedata.category(chr(i)).startswith("P")
     )

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:
                 ```
@@ -47,7 +47,7 @@ class ImageEmbedding(ImageEmbeddingBase):
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
         cuda: bool = False,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         lazy_load: bool = False,
         **kwargs,
     ):

fastembed/image/onnx_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
 
@@ -53,6 +53,17 @@ supported_onnx_models = [
         },
         "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",
+    },
 ]
 
 
@@ -64,7 +75,7 @@ class OnnxImageEmbedding(ImageEmbeddingBase, OnnxImageModel[np.ndarray]):
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
         cuda: bool = False,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         lazy_load: bool = False,
         device_id: Optional[int] = None,
         **kwargs,
@@ -80,7 +91,7 @@ class OnnxImageEmbedding(ImageEmbeddingBase, OnnxImageModel[np.ndarray]):
                 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
+            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.
@@ -129,12 +140,12 @@ class OnnxImageEmbedding(ImageEmbeddingBase, OnnxImageModel[np.ndarray]):
         )
 
     @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
 
@@ -178,8 +189,8 @@ 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
+    ) -> dict[str, np.ndarray]:
         """
         Preprocess the onnx input.
         """

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
@@ -29,8 +29,8 @@ class OnnxImageModel(OnnxModel[T]):
         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
+    ) -> dict[str, np.ndarray]:
         """
         Preprocess the onnx input.
         """
@@ -58,10 +58,10 @@ class OnnxImageModel(OnnxModel[T]):
     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]:
+    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) -> OnnxOutputContext:
         with contextlib.ExitStack():
             image_files = [
                 Image.open(image) if not isinstance(image, Image.Image) else image
@@ -83,7 +83,7 @@ class OnnxImageModel(OnnxModel[T]):
         parallel: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
         cuda: bool = False,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         **kwargs,
     ) -> Iterable[T]:
         is_small = False
@@ -125,7 +125,7 @@ class OnnxImageModel(OnnxModel[T]):
 
 
 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)
 
@@ -122,3 +122,29 @@ def pil2ndarray(image: Union[Image.Image, 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]):
         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]):
         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]):
         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]):
         transforms.append(PILtoNDarray())
 
     @staticmethod
-    def _get_rescale(transforms: List[Transform], config: Dict[str, Any]):
+    def _get_rescale(transforms: list[Transform], config: dict[str, Any]):
         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]):
         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]):
+        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
@@ -12,6 +12,7 @@ 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",
@@ -41,7 +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(
@@ -67,25 +68,21 @@ 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
-        # make sure that "@ " is considered as a single token
-        query = f"@ {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:
@@ -104,19 +101,16 @@ class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[np.ndarray]):
                 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
-        # make sure that "@ " is considered as a single 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
 
@@ -127,7 +121,7 @@ class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[np.ndarray]):
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
         cuda: bool = False,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         lazy_load: bool = False,
         device_id: Optional[int] = None,
         **kwargs,
@@ -143,7 +137,7 @@ class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[np.ndarray]):
                 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
+            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.
@@ -197,6 +191,9 @@ class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[np.ndarray]):
             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,
@@ -232,7 +229,7 @@ class Colbert(LateInteractionTextEmbeddingBase, OnnxTextModel[np.ndarray]):
             **kwargs,
         )
 
-    def query_embed(self, query: Union[str, List[str]], **kwargs) -> Iterable[np.ndarray]:
+    def query_embed(self, query: Union[str, Iterable[str]], **kwargs) -> Iterable[np.ndarray]:
         if isinstance(query, str):
             query = [query]
 

fastembed/late_interaction/jina_colbert.py

@@ -1,10 +1,11 @@
-from typing import Any, Dict, List, Type
+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",
@@ -24,7 +25,7 @@ supported_jina_colbert_models = [
 class JinaColbert(Colbert):
     QUERY_MARKER_TOKEN_ID = 250002
     DOCUMENT_MARKER_TOKEN_ID = 250003
-    MIN_QUERY_LENGTH = 32
+    MIN_QUERY_LENGTH = 31  # it's 32, we add one additional special token in the beginning
     MASK_TOKEN = "<mask>"
 
     @classmethod
@@ -32,22 +33,21 @@ class JinaColbert(Colbert):
         return JinaColbertEmbeddingWorker
 
     @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_jina_colbert_models
 
     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
-            # the attention mask for jina-colbert-v2 is always 1 in queries
+        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
 

fastembed/late_interaction/late_interaction_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
 
 import numpy as np
 
@@ -11,15 +11,15 @@ from fastembed.late_interaction.late_interaction_embedding_base import (
 
 
 class LateInteractionTextEmbedding(LateInteractionTextEmbeddingBase):
-    EMBEDDINGS_REGISTRY: List[Type[LateInteractionTextEmbeddingBase]] = [Colbert, JinaColbert]
+    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:
                 ```
@@ -50,7 +50,7 @@ class LateInteractionTextEmbedding(LateInteractionTextEmbeddingBase):
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
         cuda: bool = False,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         lazy_load: bool = False,
         **kwargs,
     ):

fastembed/parallel_processor.py

@@ -1,14 +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 copy import deepcopy
+from typing import Any, Iterable, Optional, Type
+
 
 # Single item should be processed in less than:
 processing_timeout = 10 * 60  # seconds
@@ -24,10 +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()
 
 
@@ -37,7 +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.
@@ -93,7 +94,7 @@ class ParallelWorkerPool:
         num_workers: int,
         worker: Type[Worker],
         start_method: Optional[str] = None,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         cuda: bool = False,
     ):
         self.worker_class = worker
@@ -101,7 +102,7 @@ class ParallelWorkerPool:
         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
@@ -150,7 +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)
 

fastembed/rerank/cross_encoder/onnx_text_cross_encoder.py

@@ -1,11 +1,15 @@
-from typing import List, Iterable, Dict, Any, Sequence, Optional
+from typing import Any, Iterable, Optional, Sequence, Type
 
 from loguru import logger
 
 from fastembed.common import OnnxProvider
-from fastembed.rerank.cross_encoder.onnx_text_model import OnnxCrossEncoderModel
-from fastembed.rerank.cross_encoder.text_cross_encoder_base import TextCrossEncoderBase
+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 = [
     {
@@ -38,16 +42,46 @@ supported_onnx_models = [
         "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]]:
+    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
 
@@ -58,10 +92,10 @@ class OnnxTextCrossEncoder(TextCrossEncoderBase, OnnxCrossEncoderModel):
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
         cuda: bool = False,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         lazy_load: bool = False,
         device_id: Optional[int] = None,
-        **kwargs,
+        **kwargs: Any,
     ):
         """
         Args:
@@ -74,7 +108,7 @@ class OnnxTextCrossEncoder(TextCrossEncoderBase, OnnxCrossEncoderModel):
                 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
+            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.
@@ -108,7 +142,9 @@ class OnnxTextCrossEncoder(TextCrossEncoderBase, OnnxCrossEncoderModel):
         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
+            self.model_description,
+            self.cache_dir,
+            local_files_only=self._local_files_only,
         )
 
         if not self.lazy_load:
@@ -129,7 +165,7 @@ class OnnxTextCrossEncoder(TextCrossEncoderBase, OnnxCrossEncoderModel):
         query: str,
         documents: Iterable[str],
         batch_size: int = 64,
-        **kwargs,
+        **kwargs: Any,
     ) -> Iterable[float]:
         """Reranks documents based on their relevance to a given query.
 
@@ -145,3 +181,44 @@ class OnnxTextCrossEncoder(TextCrossEncoderBase, OnnxCrossEncoderModel):
         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,
+    ) -> OnnxTextCrossEncoder:
+        return OnnxTextCrossEncoder(
+            model_name=model_name,
+            cache_dir=cache_dir,
+            threads=1,
+            **kwargs,
+        )

fastembed/rerank/cross_encoder/onnx_text_model.py

@@ -1,16 +1,28 @@
-from typing import Sequence, Optional, List, Dict, Iterable
+import os
+from multiprocessing import get_all_start_methods
 from pathlib import Path
+from typing import Any, Iterable, Optional, Sequence, Type
 
 import numpy as np
 from tokenizers import Encoding
 
-from fastembed.common.onnx_model import OnnxModel, OnnxProvider
+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):
-    ONNX_OUTPUT_NAMES: Optional[List[str]] = None
+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,
@@ -31,40 +43,108 @@ class OnnxCrossEncoderModel(OnnxModel):
         )
         self.tokenizer, _ = load_tokenizer(model_dir=model_dir)
 
-    def tokenize(self, query: str, documents: List[str], **kwargs) -> List[Encoding]:
-        return self.tokenizer.encode_batch([(query, doc) for doc in documents])
-
-    def onnx_embed(self, query: str, documents: List[str], **kwargs) -> List[float]:
-        tokenized_input = self.tokenize(query, documents, **kwargs)
+    def tokenize(self, pairs: list[tuple[str, str]], **_: Any) -> list[Encoding]:
+        return self.tokenizer.encode_batch(pairs)
 
+    def _build_onnx_input(self, tokenized_input):
+        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),
-            "attention_mask": np.array(
-                [enc.attention_mask for enc in tokenized_input], dtype=np.int64
-            ),
         }
-        input_names = {node.name for node in self.model.get_inputs()}
         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)
-        return outputs[0][:, 0].tolist()
+        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
+        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.onnx_embed(query, batch, **kwargs)
+            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
-    ) -> Dict[str, np.ndarray]:
+        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

@@ -1,21 +1,21 @@
-from typing import Any, Dict, Iterable, List, Optional, Sequence, Type
+from typing import Any, Iterable, Optional, Sequence, Type
 
-from fastembed.rerank.cross_encoder.text_cross_encoder_base import TextCrossEncoderBase
-from fastembed.rerank.cross_encoder.onnx_text_cross_encoder import OnnxTextCrossEncoder
 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]] = [
+    CROSS_ENCODER_REGISTRY: list[Type[TextCrossEncoderBase]] = [
         OnnxTextCrossEncoder,
     ]
 
     @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:
                 ```
@@ -45,9 +45,9 @@ class TextCrossEncoder(TextCrossEncoderBase):
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
         cuda: bool = False,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         lazy_load: bool = False,
-        **kwargs,
+        **kwargs: Any,
     ):
         super().__init__(model_name, cache_dir, threads, **kwargs)
 
@@ -72,7 +72,7 @@ class TextCrossEncoder(TextCrossEncoderBase):
         )
 
     def rerank(
-        self, query: str, documents: Iterable[str], batch_size: int = 64, **kwargs
+        self, query: str, documents: Iterable[str], batch_size: int = 64, **kwargs: Any
     ) -> Iterable[float]:
         """Rerank a list of documents based on a query.
 
@@ -85,3 +85,36 @@ class TextCrossEncoder(TextCrossEncoderBase):
             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

@@ -1,4 +1,4 @@
-from typing import Iterable, Optional
+from typing import Any, Iterable, Optional
 
 from fastembed.common.model_management import ModelManagement
 
@@ -23,7 +23,7 @@ class TextCrossEncoderBase(ModelManagement):
         batch_size: int = 64,
         **kwargs,
     ) -> Iterable[float]:
-        """Reranks a list of documents given a query.
+        """Rerank a list of documents given a query.
 
         Args:
             query (str): The query to rerank the documents.
@@ -32,6 +32,27 @@ class TextCrossEncoderBase(ModelManagement):
             **kwargs: Additional keyword argument to pass to the rerank method.
 
         Yields:
-            Iterable[float]: The scores of reranked the documents.
+            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

@@ -2,7 +2,7 @@ import os
 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
@@ -92,6 +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.
     """
@@ -105,6 +107,7 @@ class Bm25(SparseTextEmbeddingBase):
         avg_len: float = 256.0,
         language: str = "english",
         token_max_length: int = 40,
+        disable_stemmer: bool = False,
         **kwargs,
     ):
         super().__init__(model_name, cache_dir, **kwargs)
@@ -127,22 +130,28 @@ class Bm25(SparseTextEmbeddingBase):
 
         self.token_max_length = token_max_length
         self.punctuation = set(get_all_punctuation())
-        self.stopwords = set(self._load_stopwords(self._model_dir, self.language))
+        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.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 []
@@ -182,6 +191,9 @@ 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(
                 num_workers=parallel or 1,
@@ -222,19 +234,21 @@ 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.lower() in self.stopwords:
+            if lower_token in self.stopwords:
                 continue
 
             if len(token) > self.token_max_length:
                 continue
 
-            stemmed_token = self.stemmer.stem_word(token.lower())
+            stemmed_token = self.stemmer.stem_word(lower_token) if self.stemmer else lower_token
 
             if stemmed_token:
                 stemmed_tokens.append(stemmed_token)
@@ -242,8 +256,8 @@ 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)
@@ -253,7 +267,7 @@ class Bm25(SparseTextEmbeddingBase):
             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.
 
         (
@@ -263,10 +277,10 @@ 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)
@@ -323,7 +337,7 @@ 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

fastembed/sparse/bm42.py

@@ -1,7 +1,7 @@
 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
@@ -63,7 +63,7 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
         providers: Optional[Sequence[OnnxProvider]] = None,
         alpha: float = 0.5,
         cuda: bool = False,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         lazy_load: bool = False,
         device_id: Optional[int] = None,
         **kwargs,
@@ -81,7 +81,7 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
                 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
+            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.
@@ -141,7 +141,7 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
         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]]:
+    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:
@@ -149,7 +149,7 @@ 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.stem_word(token)
@@ -158,8 +158,8 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
 
     @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)
@@ -167,8 +167,8 @@ 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 = []
@@ -195,7 +195,7 @@ 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.
@@ -246,16 +246,16 @@ 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 []
@@ -298,7 +298,7 @@ class Bm42(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
         )
 
     @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))

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
 
 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())
@@ -50,9 +50,7 @@ class SparseTextEmbeddingBase(ModelManagement):
     ) -> Iterable[SparseEmbedding]:
         raise NotImplementedError()
 
-    def passage_embed(
-        self, texts: Iterable[str], **kwargs
-    ) -> Iterable[SparseEmbedding]:
+    def passage_embed(self, texts: Iterable[str], **kwargs) -> Iterable[SparseEmbedding]:
         """
         Embeds a list of text passages into a list of embeddings.
 
@@ -67,9 +65,7 @@ class SparseTextEmbeddingBase(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[SparseEmbedding]:
+    def query_embed(self, query: Union[str, Iterable[str]], **kwargs) -> 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:
                 ```
@@ -50,7 +50,7 @@ class SparseTextEmbedding(SparseTextEmbeddingBase):
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
         cuda: bool = False,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         lazy_load: bool = False,
         **kwargs,
     ):

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
@@ -55,11 +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
 
@@ -70,7 +70,7 @@ class SpladePP(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
         cuda: bool = False,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         lazy_load: bool = False,
         device_id: Optional[int] = None,
         **kwargs,
@@ -86,7 +86,7 @@ class SpladePP(SparseTextEmbeddingBase, OnnxTextModel[SparseEmbedding]):
                 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
+            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.

fastembed/sparse/utils/tokenizer.py

@@ -1,11 +1,11 @@
 # This code is a modified copy of the `NLTKWordTokenizer` class from `NLTK` library.
 
 import re
-from typing import List
 
 
 class SimpleTokenizer:
-    def tokenize(text: str) -> List[str]:
+    @staticmethod
+    def tokenize(text: str) -> list[str]:
         text = re.sub(r"[^\w]", " ", text.lower())
         text = re.sub(r"\s+", " ", text)
 
@@ -80,7 +80,7 @@ class WordTokenizer:
     ]
 
     @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
 
@@ -27,11 +27,11 @@ 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
 

fastembed/text/e5_onnx_embedding.py

@@ -1,4 +1,4 @@
-from typing import Any, Dict, List, Type
+from typing import Any, Type
 
 import numpy as np
 
@@ -39,17 +39,17 @@ class E5OnnxEmbedding(OnnxTextEmbedding):
         return E5OnnxEmbeddingWorker
 
     @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_multilingual_e5_models
 
     def _preprocess_onnx_input(
-        self, onnx_input: Dict[str, np.ndarray], **kwargs
-    ) -> Dict[str, np.ndarray]:
+        self, onnx_input: dict[str, np.ndarray], **kwargs
+    ) -> dict[str, np.ndarray]:
         """
         Preprocess the onnx input.
         """

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
 
@@ -16,6 +16,7 @@ supported_onnx_models = [
         "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",
@@ -50,6 +51,7 @@ supported_onnx_models = [
         "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",
@@ -72,6 +74,7 @@ supported_onnx_models = [
         "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",
@@ -164,6 +167,17 @@ supported_onnx_models = [
         },
         "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",
+    },
 ]
 
 
@@ -171,12 +185,12 @@ 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
 
@@ -187,7 +201,7 @@ class OnnxTextEmbedding(TextEmbeddingBase, OnnxTextModel[np.ndarray]):
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
         cuda: bool = False,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         lazy_load: bool = False,
         device_id: Optional[int] = None,
         **kwargs,
@@ -203,7 +217,7 @@ class OnnxTextEmbedding(TextEmbeddingBase, OnnxTextModel[np.ndarray]):
                 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
+            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.
@@ -276,8 +290,8 @@ 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
+    ) -> dict[str, np.ndarray]:
         """
         Preprocess the onnx input.
         """
@@ -285,7 +299,13 @@ 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(

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"]:
@@ -29,8 +29,8 @@ class OnnxTextModel(OnnxModel[T]):
         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
+    ) -> dict[str, np.ndarray]:
         """
         Preprocess the onnx input.
         """
@@ -58,12 +58,12 @@ class OnnxTextModel(OnnxModel[T]):
     def load_onnx_model(self) -> None:
         raise NotImplementedError("Subclasses must implement this method")
 
-    def tokenize(self, documents: List[str], **kwargs) -> List[Encoding]:
+    def tokenize(self, documents: list[str], **kwargs) -> list[Encoding]:
         return self.tokenizer.encode_batch(documents)
 
     def onnx_embed(
         self,
-        documents: List[str],
+        documents: list[str],
         **kwargs,
     ) -> OnnxOutputContext:
         encoded = self.tokenize(documents, **kwargs)
@@ -79,7 +79,6 @@ 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)
@@ -98,7 +97,7 @@ class OnnxTextModel(OnnxModel[T]):
         parallel: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
         cuda: bool = False,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         **kwargs,
     ) -> Iterable[T]:
         is_small = False
@@ -140,7 +139,7 @@ class OnnxTextModel(OnnxModel[T]):
 
 
 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

@@ -1,4 +1,4 @@
-from typing import Any, Dict, Iterable, List, Type
+from typing import Any, Iterable, Type
 
 import numpy as np
 
@@ -60,11 +60,11 @@ 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
 

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
 
@@ -57,6 +57,33 @@ supported_pooled_normalized_models = [
         "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",
+    },
 ]
 
 
@@ -66,11 +93,11 @@ 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
 

fastembed/text/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
 
 import numpy as np
 
@@ -12,7 +12,7 @@ from fastembed.text.text_embedding_base import TextEmbeddingBase
 
 
 class TextEmbedding(TextEmbeddingBase):
-    EMBEDDINGS_REGISTRY: List[Type[TextEmbeddingBase]] = [
+    EMBEDDINGS_REGISTRY: list[Type[TextEmbeddingBase]] = [
         OnnxTextEmbedding,
         E5OnnxEmbedding,
         CLIPOnnxEmbedding,
@@ -21,12 +21,12 @@ class TextEmbedding(TextEmbeddingBase):
     ]
 
     @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:
                 ```
@@ -57,7 +57,7 @@ class TextEmbedding(TextEmbeddingBase):
         threads: Optional[int] = None,
         providers: Optional[Sequence[OnnxProvider]] = None,
         cuda: bool = False,
-        device_ids: Optional[List[int]] = None,
+        device_ids: Optional[list[int]] = None,
         lazy_load: bool = False,
         **kwargs,
     ):
@@ -78,7 +78,7 @@ class TextEmbedding(TextEmbeddingBase):
                 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()`"
         )
 

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "fastembed"
-version = "0.4.2"
+version = "0.5.0"
 description = "Fast, light, accurate library built for retrieval embedding generation"
 authors = ["Qdrant Team <info@qdrant.tech>", "NirantK <nirant.bits@gmail.com>"]
 license = "Apache License"
@@ -11,18 +11,24 @@ 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,<1.20.0"
+python = ">=3.9.0"
+onnx = ">=1.15.0"
+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"
 mmh3 = "^4.1.0"
 py-rust-stemmers = "^0.1.0"
@@ -31,7 +37,7 @@ py-rust-stemmers = "^0.1.0"
 pytest = "^7.4.2"
 ruff = ">=0.3.1,<1.0"
 notebook = ">=7.0.2"
-pre-commit = {version = "^3.6.2", python = ">=3.9,<3.12" }
+pre-commit = "^3.6.2"
 
 [tool.poetry.group.docs.dependencies]
 mkdocs-material = "^9.5.10"

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
@@ -111,8 +111,8 @@ 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

@@ -1,10 +1,10 @@
 import os
-import shutil
 
 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"])
@@ -67,7 +67,7 @@ def test_attention_embeddings(model_name):
         assert len(result.indices) == 2
 
     if is_ci:
-        shutil.rmtree(model.model._model_dir)
+        delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize("model_name", ["Qdrant/bm42-all-minilm-l6-v2-attentions", "Qdrant/bm25"])
@@ -92,7 +92,7 @@ def test_parallel_processing(model_name):
         assert np.allclose(emb_1.values, emb_3.values)
 
     if is_ci:
-        shutil.rmtree(model.model._model_dir)
+        delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize("model_name", ["Qdrant/bm25"])
@@ -118,7 +118,7 @@ def test_multilanguage(model_name):
     assert embeddings[1].indices.shape == (4,)
 
     if is_ci:
-        shutil.rmtree(model.model._model_dir)
+        delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize("model_name", ["Qdrant/bm25"])
@@ -141,7 +141,7 @@ def test_special_characters(model_name):
         assert embeddings[idx].indices.shape == (shape,)
 
     if is_ci:
-        shutil.rmtree(model.model._model_dir)
+        delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize("model_name", ["Qdrant/bm42-all-minilm-l6-v2-attentions"])

tests/test_image_onnx_embeddings.py

@@ -1,5 +1,4 @@
 import os
-import shutil
 from io import BytesIO
 
 import numpy as np
@@ -9,6 +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]),
@@ -21,6 +21,9 @@ 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]
+    ),
 }
 
 
@@ -54,7 +57,7 @@ def test_embedding():
         assert np.allclose(embeddings[1], embeddings[2]), model_desc["model"]
 
         if is_ci:
-            shutil.rmtree(model.model._model_dir)
+            delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize("n_dims,model_name", [(512, "Qdrant/clip-ViT-B-32-vision")])
@@ -74,7 +77,7 @@ def test_batch_embedding(n_dims, model_name):
 
     assert embeddings.shape == (len(test_images) * n_images, n_dims)
     if is_ci:
-        shutil.rmtree(model.model._model_dir)
+        delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize("n_dims,model_name", [(512, "Qdrant/clip-ViT-B-32-vision")])
@@ -102,11 +105,12 @@ def test_parallel_processing(n_dims, model_name):
     assert np.allclose(embeddings, embeddings_2, atol=1e-3)
     assert np.allclose(embeddings, embeddings_3, atol=1e-3)
     if is_ci:
-        shutil.rmtree(model.model._model_dir)
+        delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize("model_name", ["Qdrant/clip-ViT-B-32-vision"])
 def test_lazy_load(model_name):
+    is_ci = os.getenv("CI")
     model = ImageEmbedding(model_name=model_name, lazy_load=True)
     assert not hasattr(model.model, "model")
     images = [
@@ -115,3 +119,5 @@ def test_lazy_load(model_name):
     ]
     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

@@ -1,5 +1,4 @@
 import os
-import shutil
 
 import pytest
 import numpy as np
@@ -7,6 +6,7 @@ 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 = {
@@ -167,7 +167,7 @@ def test_batch_embedding():
             assert np.allclose(value[:, :abridged_dim], expected_result, atol=2e-3)
 
         if is_ci:
-            shutil.rmtree(model.model._model_dir)
+            delete_model_cache(model.model._model_dir)
 
 
 def test_single_embedding():
@@ -182,7 +182,7 @@ def test_single_embedding():
         assert np.allclose(result[:, :abridged_dim], expected_result, atol=2e-3)
 
         if is_ci:
-            shutil.rmtree(model.model._model_dir)
+            delete_model_cache(model.model._model_dir)
 
 
 def test_single_embedding_query():
@@ -197,7 +197,7 @@ def test_single_embedding_query():
         assert np.allclose(result[:, :abridged_dim], expected_result, atol=2e-3)
 
         if is_ci:
-            shutil.rmtree(model.model._model_dir)
+            delete_model_cache(model.model._model_dir)
 
 
 def test_parallel_processing():
@@ -219,7 +219,7 @@ def test_parallel_processing():
     assert np.allclose(embeddings, embeddings_3, atol=1e-3)
 
     if is_ci:
-        shutil.rmtree(model.model._model_dir)
+        delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize(
@@ -227,6 +227,8 @@ def test_parallel_processing():
     ["colbert-ir/colbertv2.0"],
 )
 def test_lazy_load(model_name):
+    is_ci = os.getenv("CI")
+
     model = LateInteractionTextEmbedding(model_name=model_name, lazy_load=True)
     assert not hasattr(model.model, "model")
 
@@ -239,3 +241,6 @@ def test_lazy_load(model_name):
 
     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_sparse_embeddings.py

@@ -1,11 +1,11 @@
 import os
-import shutil
 
 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": {
@@ -61,7 +61,7 @@ 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:
-            shutil.rmtree(model.model._model_dir)
+            delete_model_cache(model.model._model_dir)
 
 
 def test_single_embedding():
@@ -77,7 +77,7 @@ 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:
-            shutil.rmtree(model.model._model_dir)
+            delete_model_cache(model.model._model_dir)
 
 
 def test_parallel_processing():
@@ -107,7 +107,7 @@ def test_parallel_processing():
         assert np.allclose(sparse_embedding.values, sparse_embedding_all.values, atol=1e-3)
 
     if is_ci:
-        shutil.rmtree(model.model._model_dir)
+        delete_model_cache(model.model._model_dir)
 
 
 @pytest.fixture
@@ -116,7 +116,7 @@ def bm25_instance():
     model = Bm25("Qdrant/bm25", language="english")
     yield model
     if ci:
-        shutil.rmtree(model._model_dir)
+        delete_model_cache(model._model_dir)
 
 
 def test_stem_with_stopwords_and_punctuation(bm25_instance):
@@ -150,11 +150,34 @@ def test_stem_case_insensitive_stopwords(bm25_instance):
     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):
+    # 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):
+    is_ci = os.getenv("CI")
     model = SparseTextEmbedding(model_name=model_name, lazy_load=True)
     assert not hasattr(model.model, "model")
 
@@ -167,3 +190,6 @@ def test_lazy_load(model_name):
 
     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

@@ -2,42 +2,56 @@ import os
 
 import numpy as np
 import pytest
-import shutil
 
 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]),
 }
 
-
-def test_rerank():
-    is_ci = os.getenv("CI")
-
-    for model_desc in TextCrossEncoder.list_supported_models():
-        if not is_ci and model_desc["size_in_GB"] > 1:
-            continue
-
-        model_name = model_desc["model"]
-        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)))
-
-        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:
-            shutil.rmtree(model.model._model_dir)
+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",
-    ["Xenova/ms-marco-MiniLM-L-6-v2", "Xenova/ms-marco-MiniLM-L-12-v2", "BAAI/bge-reranker-base"],
+    [model_name for model_name in CANONICAL_SCORE_VALUES],
+)
+def test_rerank(model_name):
+    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):
     is_ci = os.getenv("CI")
@@ -48,6 +62,12 @@ def test_batch_rerank(model_name):
     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}"
@@ -55,7 +75,7 @@ def test_batch_rerank(model_name):
         scores, canonical_scores, atol=1e-3
     ), f"Model: {model_name}, Scores: {scores}, Expected: {canonical_scores}"
     if is_ci:
-        shutil.rmtree(model.model._model_dir)
+        delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize(
@@ -63,9 +83,37 @@ def test_batch_rerank(model_name):
     ["Xenova/ms-marco-MiniLM-L-6-v2"],
 )
 def test_lazy_load(model_name):
+    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):
+    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_onnx_embeddings.py

@@ -1,10 +1,10 @@
 import os
-import shutil
 
 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]),
@@ -41,6 +41,8 @@ CANONICAL_VECTOR_VALUES = {
     "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]
@@ -62,6 +64,8 @@ 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]),
 }
 
 
@@ -85,7 +89,7 @@ def test_embedding():
             embeddings[0, : canonical_vector.shape[0]], canonical_vector, atol=1e-3
         ), model_desc["model"]
         if is_ci:
-            shutil.rmtree(model.model._model_dir)
+            delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize(
@@ -102,7 +106,7 @@ def test_batch_embedding(n_dims, model_name):
 
     assert embeddings.shape == (200, n_dims)
     if is_ci:
-        shutil.rmtree(model.model._model_dir)
+        delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize(
@@ -128,7 +132,7 @@ def test_parallel_processing(n_dims, model_name):
     assert np.allclose(embeddings, embeddings_3, atol=1e-3)
 
     if is_ci:
-        shutil.rmtree(model.model._model_dir)
+        delete_model_cache(model.model._model_dir)
 
 
 @pytest.mark.parametrize(
@@ -136,6 +140,7 @@ def test_parallel_processing(n_dims, model_name):
     ["BAAI/bge-small-en-v1.5"],
 )
 def test_lazy_load(model_name):
+    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"]
@@ -147,3 +152,6 @@ def test_lazy_load(model_name):
 
     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 +1,32 @@
+import shutil
+import traceback
+
+from pathlib import Path
+from typing import Union
+
+
+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, path, exc_info):
+        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)