Upload 11 files

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+faiss_ubinary.index filter=lfs diff=lfs merge=lfs -text
+usearch_int8.index filter=lfs diff=lfs merge=lfs -text

LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

README.md

@@ -1,13 +1,61 @@
 ---
-title: Legalkit Retrieval
-emoji: 📊
-colorFrom: red
-colorTo: yellow
+title: LegalKit Retrieval
+emoji: 📖
+colorFrom: blue
+colorTo: purple
 sdk: gradio
 sdk_version: 4.25.0
 app_file: app.py
-pinned: false
+pinned: true
 license: apache-2.0
+short_description: A binary Search with Scalar Rescoring through legal codes
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# LegalKit Retrieval, a binary Search with Scalar (int8) Rescoring through French legal codes
+
+This space showcases the [tsdae-lemone-mbert-base](https://huggingface.co/louisbrulenaudet/tsdae-lemone-mbert-base) 
+model by Louis Brulé Naudet, a sentence embedding model based on BERT fitted using Transformer-based Sequential Denoising Auto-Encoder for unsupervised sentence embedding learning with one objective : french legal domain adaptation.
+
+This process is designed to be memory efficient and fast, with the binary index being small enough to fit in memory and the int8 index being loaded as a view to save memory. 
+In total, this process requires keeping 1) the model in memory, 2) the binary index in memory, and 3) the int8 index on disk.
+
+Additionally, the binary index is much faster (up to 32x) to search than the float32 index, while the rescoring is also extremely efficient. In conclusion, this process allows for fast, scalable, cheap, and memory-efficient retrieval.
+
+Notes:
+- The SentenceTransformer model currently in use is in beta and may not be suitable for direct use in production.
+
+## Dependencies
+### Libraries Used:
+
+- **Accelerate** (v0.29.1): A Python library for high-performance computing, enabling faster execution of computational tasks.
+- **Faiss-GPU** (v1.7.2): A GPU-accelerated library for efficient similarity search and clustering of dense vectors, essential for high-dimensional data analysis.
+- **Gradio** (v4.25.0): An intuitive library for creating customizable UI components around machine learning models, simplifying model deployment and interaction.
+- **Polars** (v0.20.18): A blazing-fast DataFrame library for Rust, providing efficient data manipulation capabilities for large datasets.
+- **Sentence-Transformers** (v2.6.1): A versatile library for generating sentence embeddings, facilitating various natural language processing tasks such as semantic similarity and text classification.
+- **Spaces** (v0.25.0): A utility library designed to optimize GPU resource management, enhancing efficiency and scalability in GPU-based computing environments.
+- **Usearch** (v2.10.5): A powerful library for performing fast approximate nearest neighbor search, crucial for tasks like recommendation systems and data clustering.
+
+### Installation Guide
+
+To install all the dependencies, you can use the following command:
+
+```shell
+pip3 install accelerate faiss-gpu gradio polars sentence-transformers spaces usearch
+```
+
+Note: Ensure you have Python installed on your system before proceeding with the installation of these libraries.
+
+## Citing this project
+If you use this code in your research, please use the following BibTeX entry.
+
+```BibTeX
+@misc{louisbrulenaudet2024,
+	author = {Louis Brulé Naudet},
+	title = {LegalKit Retrieval, a binary Search with Scalar (int8) Rescoring through French legal codes},
+	howpublished = {\url{https://huggingface.co/spaces/louisbrulenaudet/legalkit-retrieval}},
+	year = {2024}
+}
+
+```
+## Feedback
+If you have any feedback, please reach out at [louisbrulenaudet@icloud.com](mailto:louisbrulenaudet@icloud.com).

app.py

@@ -0,0 +1,271 @@
+# -*- coding: utf-8 -*-
+# Copyright (c) Louis Brulé Naudet. All Rights Reserved.
+# This software may be used and distributed according to the terms of the License Agreement.
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import gradio as gr
+import polars as pl
+import spaces
+import torch
+
+from typing import Tuple, List, Union
+
+from dataset import Dataset
+from similarity_search import SimilaritySearch
+
+
+def setup(
+    description: str, 
+    model_name: str,
+    device: str,
+    ndim: int,
+    metric: str,
+    dtype: str
+) -> Tuple:
+    """
+    Set up the model and tokenizer for a given pre-trained model ID.
+
+    Parameters
+    ----------
+    description : str
+        A string containing additional description information.
+
+    model_name : str
+        Name of the pre-trained model to load.
+
+    device : str
+        Device to run the model on, e.g., 'cuda' or 'cpu'.
+
+    ndim : int
+        Dimensionality of the model.
+
+    metric : str
+        Metric for similarity search.
+
+    dtype : str
+        Data type of the model.
+
+    Returns
+    -------
+    instance : SimilaritySearch
+         A class dedicated to encoding text data, quantizing embeddings, and managing indices for efficient similarity search.
+
+    dataset : datasets.Dataset
+            The loaded dataset.
+
+    dataframe: pl.DataFrame
+            A Polars DataFrame representing the dataset.
+
+    description : str
+        A string containing additional description information.
+    """
+    try:
+        if not torch.cuda.is_available():
+            description += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
+            
+        instance = SimilaritySearch(
+            model_name=model_name,
+            device=device,
+            ndim=ndim,
+            metric=metric,
+            dtype=dtype
+        )
+
+        instance.load_usearch_index_view(
+            index_path="./usearch_int8.index",
+        )
+
+        instance.load_faiss_index(
+            index_path="./faiss_ubinary.index",
+        )
+
+        dataset = Dataset.load(
+            dataset_path="./legalkit.hf"
+        )
+
+        dataframe = Dataset.convert_to_polars(
+            dataset=dataset
+        )
+
+        return instance, dataset, dataframe, description
+
+    except Exception as e:
+        error_message = f"An error occurred during setup: {str(e)}"
+        raise RuntimeError(error_message) from e
+
+
+DESCRIPTION = """\
+# LegalKit Retrieval, a binary Search with Scalar (int8) Rescoring through French legal codes
+
+This space showcases the [tsdae-lemone-mbert-base](https://huggingface.co/louisbrulenaudet/tsdae-lemone-mbert-base) 
+model by Louis Brulé Naudet, a sentence embedding model based on BERT fitted using Transformer-based Sequential Denoising Auto-Encoder for unsupervised sentence embedding learning with one objective : french legal domain adaptation.
+
+This process is designed to be memory efficient and fast, with the binary index being small enough to fit in memory and the int8 index being loaded as a view to save memory. 
+Additionally, the binary index is much faster (up to 32x) to search than the float32 index, while the rescoring is also extremely efficient.
+"""
+
+instance, dataset, dataframe, DESCRIPTION = setup(
+    model_name="louisbrulenaudet/tsdae-lemone-mbert-base",
+    description=DESCRIPTION,
+    device="cuda",
+    ndim=768,
+    metric="ip",
+    dtype="i8"
+)
+
+
+@spaces.GPU
+def search(
+    query:str, 
+    top_k:int, 
+    rescore_multiplier:int
+) -> any:
+    """
+    Perform a search operation using the initialized GPU space.
+
+    Parameters
+    ----------
+    query : str
+        The query for which similarity search is performed.
+
+    top_k : int
+        The number of top results to return.
+
+    rescore_multiplier : int
+        A multiplier for rescore operation.
+
+    Returns
+    -------
+    any
+        The search results in a suitable format.
+
+    Notes
+    -----
+    This function performs a search operation using the initialized GPU space
+    and returns the search results in a format compatible with the provided
+    space.
+
+    Examples
+    --------
+    >>> results = search(query="example query", top_k=10, rescore_multiplier=2)
+    """
+    global instance
+    global dataset
+    global dataframe
+
+    top_k_scores, top_k_indices = instance.search(
+        query=query,
+        top_k=top_k,
+        rescore_multiplier=rescore_multiplier
+    )
+
+    scores_df = pl.DataFrame(
+        {
+            "index": top_k_indices, 
+            "score": top_k_scores
+        }
+    ).with_columns(
+        pl.col("index").cast(pl.UInt32)
+    )
+
+    results_df = dataframe.filter(
+        pl.col("index").is_in(top_k_indices)
+    ).join(
+        scores_df,
+        how="inner",
+        on="index"
+    ).sort(
+        by="score",
+        descending=True
+    ).select(
+        [
+            "score",
+            "input",
+            "output",
+            "start",
+            "expiration"
+        ]
+    )
+
+    return gr.Dataframe(
+        value=results_df,
+        type="polars",
+        render=True
+    )
+
+
+with gr.Blocks(title="Quantized Retrieval") as demo:
+    gr.Markdown(
+        value=DESCRIPTION
+    )
+    gr.DuplicateButton()
+
+    with gr.Row():
+        with gr.Column():
+            query = gr.Textbox(
+                label="Query for French legal codes articles",
+                placeholder="Enter a query to search for relevant texts from the French law.",
+            )
+
+    with gr.Row():
+        with gr.Column(scale=2):
+            top_k = gr.Slider(
+                minimum=1,
+                maximum=100,
+                step=1,
+                value=20,
+                label="Number of documents to retrieve",
+                info="Number of documents to retrieve from the binary search.",
+            )
+        with gr.Column(scale=2):
+            rescore_multiplier = gr.Slider(
+                minimum=1,
+                maximum=10,
+                step=1,
+                value=4,
+                label="Rescore multiplier",
+                info="Search for 'rescore_multiplier' as many documents to rescore.",
+            )
+
+    search_button = gr.Button(value="Search")
+
+    with gr.Row():
+        with gr.Column():
+            output = gr.Dataframe(
+                render=False
+            )
+
+    query.submit(
+        search, 
+        inputs=[
+            query, 
+            top_k, 
+            rescore_multiplier
+        ], 
+        outputs=[
+            output
+        ]
+    )
+
+    search_button.click(
+        search, 
+        inputs=[
+            query, 
+            top_k, 
+            rescore_multiplier
+        ], 
+        outputs=[
+            output,
+        ]
+    )
+
+
+if __name__ == "__main__":
+    demo.queue().launch(
+        show_api=False
+    )

dataset.py

@@ -0,0 +1,131 @@
+# -*- coding: utf-8 -*-
+# Copyright (c) Louis Brulé Naudet. All Rights Reserved.
+# This software may be used and distributed according to the terms of the License Agreement.
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import datasets
+import polars as pl
+
+
+class Dataset:
+    @staticmethod
+    def load(
+        dataset_path:str
+    ):
+        """
+        Load a dataset from disk.
+
+        Parameters
+        ----------
+        dataset_path : str
+            The path to the dataset on disk.
+
+        Returns
+        -------
+        datasets.Dataset
+            The loaded dataset.
+
+        Notes
+        -----
+        This method statically loads a dataset from disk using the `load_from_disk` function
+        provided by the `datasets` module. The dataset is expected to be stored in a specific
+        format supported by the `datasets` library.
+
+        Example
+        -------
+        >>> dataset_path = "/path/to/dataset"
+        >>> dataset = Dataset.load(dataset_path)
+        """
+        dataset = datasets.load_from_disk(
+            dataset_path=dataset_path
+        )
+
+        return dataset
+
+
+    @staticmethod
+    def save(
+        dataset: datasets.Dataset,
+        dataset_path: str
+    ) -> None:
+        """
+        Save a dataset to disk.
+
+        Parameters
+        ----------
+        dataset : datasets.Dataset
+            The dataset to be saved.
+
+        dataset_path : str
+            The path where the dataset will be saved on disk.
+
+        Returns
+        -------
+        None
+
+        Notes
+        -----
+        This method statically saves a dataset to disk using the `save_to_disk` function
+        provided by the `datasets` module. The dataset is expected to be in a format
+        supported by the `datasets` library.
+
+        Example
+        -------
+        >>> dataset = load_dataset("my_dataset")
+        >>> dataset_path = "/path/to/save/dataset"
+        >>> Dataset.save(dataset, dataset_path)
+        """
+        datasets.save_to_disk(
+            dataset, 
+            dataset_path
+        )
+
+        return None
+
+    @staticmethod
+    def convert_to_polars(
+        dataset: datasets.Dataset
+    ) -> pl.DataFrame:
+        """
+        Convert a dataset to a Polars DataFrame.
+
+        Parameters
+        ----------
+        dataset : datasets.Dataset
+            The dataset to be converted to a Polars DataFrame.
+
+        Returns
+        -------
+        pl.DataFrame
+            A Polars DataFrame representing the dataset.
+
+        Notes
+        -----
+        This method converts a dataset object to a Polars DataFrame, which is a
+        memory-efficient and fast data manipulation library for Rust.
+
+        Raises
+        ------
+        Exception
+            If an error occurs during the conversion process.
+
+        Examples
+        --------
+        >>> dataset = datasets.Dataset(data=arrow_table)
+        >>> dataframe = ClassName.convert_to_polars(dataset)
+        """
+        try:
+            dataframe = pl.from_arrow(dataset.data.table).with_row_index()
+
+        except:
+            dataframe = pl.from_arrow(dataset.data.table).with_row_count(
+                name="index"
+            )
+
+        return dataframe
+

faiss_ubinary.index

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7e8b15577db9edc73dcf8e2a23c500ffe0b87e15e5f70ede4f7fb4036acd344
+size 15217569

legalkit.hf/data-00000-of-00001.arrow

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac4b9fc03afc706ccff577b7740e559a21b8821ab8472b54eff549aef580c5bf
+size 161264032

legalkit.hf/dataset_info.json

@@ -0,0 +1,32 @@
+{
+  "citation": "",
+  "description": "",
+  "features": {
+    "instruction": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "input": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "output": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "start": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "expiration": {
+      "dtype": "string",
+      "_type": "Value"
+    },
+    "num": {
+      "dtype": "string",
+      "_type": "Value"
+    }
+  },
+  "homepage": "",
+  "license": ""
+}

legalkit.hf/state.json

@@ -0,0 +1,20 @@
+{
+  "_data_files": [
+    {
+      "filename": "data-00000-of-00001.arrow"
+    }
+  ],
+  "_fingerprint": "aeae96a548e712fe",
+  "_format_columns": [
+    "instruction",
+    "input",
+    "output",
+    "start",
+    "expiration",
+    "num"
+  ],
+  "_format_kwargs": {},
+  "_format_type": null,
+  "_output_all_columns": false,
+  "_split": null
+}

requirements.txt

@@ -0,0 +1,7 @@
+accelerate==0.29.1
+faiss-gpu==1.7.2
+gradio==4.25.0
+polars==0.20.18
+sentence-transformers==2.6.1
+spaces==0.25.0
+usearch==2.10.5

similarity_search.py

@@ -0,0 +1,539 @@
+# -*- coding: utf-8 -*-
+# Copyright (c) Louis Brulé Naudet. All Rights Reserved.
+# This software may be used and distributed according to the terms of the License Agreement.
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import faiss
+import numpy as np
+import torch
+
+from usearch.index import Index
+
+from sentence_transformers import SentenceTransformer
+from sentence_transformers.quantization import quantize_embeddings
+
+from typing import Tuple, List, Union
+
+class SimilaritySearch:
+    """
+    A class dedicated to encoding text data, quantizing embeddings, and managing indices for efficient similarity search.
+
+    Attributes
+    ----------
+    model_name : str
+        Name or identifier of the embedding model.
+
+    device : str
+        Computation device ('cpu' or 'cuda').
+
+    ndim : int
+        Dimension of the embeddings.
+
+    metric : str
+        Metric used for the index ('ip' for inner product, etc.).
+
+    dtype : str
+        Data type for the index ('i8' for int8, etc.).
+
+    Methods
+    -------
+    encode(corpus, normalize_embeddings=True)
+        Encodes a list of text data into embeddings.
+
+    quantize_embeddings(embeddings, quantization_type)
+        Quantizes the embeddings for efficient storage and search.
+
+    create_faiss_index(ubinary_embeddings, index_path)
+        Creates and saves a FAISS binary index.
+
+    create_usearch_index(int8_embeddings, index_path)
+        Creates and saves a USEARCH integer index.
+
+    load_usearch_index_view(index_path)
+        Loads a USEARCH index as a view for memory-efficient operations.
+
+    load_faiss_index(index_path)
+        Loads a FAISS binary index for searching.
+
+    search(query, top_k=10, rescore_multiplier=4)
+        Performs a search operation against the indexed embeddings.
+    """
+    def __init__(
+        self,
+        model_name: str,
+        device: str = "cuda",
+        ndim: int = 1024,
+        metric: str = "ip",
+        dtype: str = "i8"
+    ):
+        """
+        Initializes the EmbeddingIndexer with the specified model, device, and index configurations.
+
+        Parameters
+        ----------
+        model_name : str
+            The name or identifier of the SentenceTransformer model to use for embedding.
+
+        device : str, optional
+            The computation device to use ('cpu' or 'cuda'). Default is 'cuda'.
+
+        ndim : int, optional
+            The dimensionality of the embeddings. Default is 1024.
+
+        metric : str, optional
+            The metric used for the index ('ip' for inner product). Default is 'ip'.
+
+        dtype : str, optional
+            The data type for the USEARCH index ('i8' for 8-bit integer). Default is 'i8'.
+        """
+        self.model_name = model_name
+        self.device = device
+        self.ndim = ndim
+        self.metric = metric
+        self.dtype = dtype
+        self.model = SentenceTransformer(
+            self.model_name,
+            device=self.device
+        )
+
+        self.binary_index = None
+        self.int8_index = None
+
+
+    def encode(
+        self,
+        corpus: list,
+        normalize_embeddings: bool = True
+    ) -> np.ndarray:
+        """
+        Encodes the given corpus into full-precision embeddings.
+
+        Parameters
+        ----------
+        corpus : list
+            A list of sentences to be encoded.
+
+        normalize_embeddings : bool, optional
+            Whether to normalize returned vectors to have length 1. In that case,
+            the faster dot-product (util.dot_score) instead of cosine similarity can be used. Default is True.
+
+        Returns
+        -------
+        np.ndarray
+            The full-precision embeddings of the corpus.
+
+        Notes
+        -----
+        This method normalizes the embeddings and shows the progress bar during the encoding process.
+        """
+        try:
+            embeddings = self.model.encode(
+                corpus,
+                normalize_embeddings=normalize_embeddings,
+                show_progress_bar=True
+            )
+            return embeddings
+
+        except Exception as e:
+            print(f"An error occurred during encoding: {e}")
+
+
+    def quantize_embeddings(
+        self,
+        embeddings: np.ndarray,
+        quantization_type: str
+    ) -> Union[np.ndarray, bytearray]:
+        """
+        Quantizes the given embeddings based on the specified quantization type ('ubinary' or 'int8').
+
+        Parameters
+        ----------
+        embeddings : np.ndarray
+            The full-precision embeddings to be quantized.
+        quantization_type : str
+            The type of quantization ('ubinary' for unsigned binary, 'int8' for 8-bit integers).
+
+        Returns
+        -------
+        Union[np.ndarray, bytearray]
+            The quantized embeddings.
+
+        Raises
+        ------
+        ValueError
+            If an unsupported quantization type is provided.
+        """
+        try:
+            if quantization_type == "ubinary":
+                return self._quantize_to_ubinary(
+                    embeddings=embeddings
+                )
+
+            elif quantization_type == "int8":
+                return self._quantize_to_int8(
+                    embeddings=embeddings
+                )
+
+            else:
+                raise ValueError(f"Unsupported quantization type: {quantization_type}")
+
+        except Exception as e:
+            print(f"An error occurred during quantization: {e}")
+
+
+    def create_faiss_index(
+        self,
+        ubinary_embeddings: bytearray,
+        index_path: str = None,
+        save: bool = False
+    ) -> None:
+        """
+        Creates and saves a FAISS binary index from ubinary embeddings.
+
+        Parameters
+        ----------
+        ubinary_embeddings : bytearray
+            The ubinary-quantized embeddings.
+
+        index_path : str, optional
+            The file path to save the FAISS binary index. Default is None.
+
+        save : bool, optional
+            Indicator for saving the index. Default is False.
+
+        Notes
+        -----
+        The dimensionality of the index is specified during the class initialization (default is 1024).
+        """
+        try:
+            self.binary_index = faiss.IndexBinaryFlat(
+                self.ndim
+            )
+            self.binary_index.add(
+                ubinary_embeddings
+            )
+
+            if save and index_path:
+                self._save_faiss_index_binary(
+                    index_path=index_path
+                )
+
+        except Exception as e:
+            print(f"An error occurred during index creation: {e}")
+
+
+    def create_usearch_index(
+        self,
+        int8_embeddings: np.ndarray,
+        index_path: str = None,
+        save: bool = False
+    ) -> None:
+        """
+        Creates and saves a USEARCH integer index from int8 embeddings.
+
+        Parameters
+        ----------
+        int8_embeddings : np.ndarray
+            The int8-quantized embeddings.
+
+        index_path : str, optional
+            The file path to save the USEARCH integer index. Default is None.
+
+        save : bool, optional
+            Indicator for saving the index. Default is False.
+
+        Returns
+        -------
+        None
+
+        Notes
+        -----
+        The dimensionality and metric of the index are specified during class initialization.
+        """
+        try:
+            self.int8_index = Index(
+                ndim=self.ndim,
+                metric=self.metric,
+                dtype=self.dtype
+            )
+
+            self.int8_index.add(
+                np.arange(
+                    len(int8_embeddings)
+                ),
+                int8_embeddings
+            )
+
+            if save == True and index_path:
+                self._save_int8_index(
+                    index_path=index_path
+                )
+
+            return self.int8_index
+
+        except Exception as e:
+            print(f"An error occurred during USEARCH index creation: {e}")
+
+
+    def load_usearch_index_view(
+        self,
+        index_path: str
+    ) -> any:
+        """
+        Loads a USEARCH index as a view for memory-efficient operations.
+
+        Parameters
+        ----------
+        index_path : str
+            The file path to the USEARCH index to be loaded as a view.
+
+        Returns
+        -------
+        object
+            A view of the USEARCH index for memory-efficient similarity search operations.
+
+        Notes
+        -----
+        Implementing this would depend on the specific USEARCH index handling library being used.
+        """
+        try:
+            self.int8_index = Index.restore(
+                index_path,
+                view=True
+            )
+
+            return self.int8_index
+
+        except Exception as e:
+            print(f"An error occurred while loading USEARCH index: {e}")
+
+
+    def load_faiss_index(
+        self,
+        index_path: str
+    ) -> None:
+        """
+        Loads a FAISS binary index from a specified file path.
+
+        This method loads a binary index created by FAISS into the class
+        attribute `binary_index`, ready for performing similarity searches.
+
+        Parameters
+        ----------
+        index_path : str
+            The file path to the saved FAISS binary index.
+
+        Returns
+        -------
+        None
+
+        Notes
+        -----
+        The loaded index is stored in the `binary_index` attribute of the class.
+        Ensure that the index at `index_path` is compatible with the configurations
+        (e.g., dimensions) used for this class instance.
+        """
+        try:
+            self.binary_index = faiss.read_index_binary(
+                index_path
+            )
+
+        except Exception as e:
+            print(f"An error occurred while loading the FAISS index: {e}")
+
+
+    def search(
+        self,
+        query: str,
+        top_k: int = 10,
+        rescore_multiplier: int = 4
+    ) -> Tuple[List[float], List[int]]:
+        """
+        Performs a search operation against the indexed embeddings.
+
+        Parameters
+        ----------
+        query : str
+            The query sentence/string to be searched.
+
+        top_k : int, optional
+            The number of top results to return.
+
+        rescore_multiplier : int, optional
+            The multiplier used to increase the initial retrieval size for re-scoring.
+            Higher values can increase precision at the cost of performance.
+
+        Returns
+        -------
+        Tuple[List[float], List[int]]
+            A tuple containing the scores and the indices of the top k results.
+
+        Notes
+        -----
+        This method assumes that `binary_index` and `int8_index` are already loaded or created.
+        """
+        try:
+            if self.binary_index is None or self.int8_index is None:
+                raise ValueError("Indices must be loaded or created before searching.")
+
+            query_embedding = self.encode(
+                corpus=query,
+                normalize_embeddings=False
+            )
+
+            query_embedding_ubinary = self.quantize_embeddings(
+                embeddings=query_embedding.reshape(1, -1),
+                quantization_type="ubinary"
+            )
+
+            _scores, binary_ids = self.binary_index.search(
+                query_embedding_ubinary,
+                top_k * rescore_multiplier
+            )
+
+            binary_ids = binary_ids[0]
+
+            int8_embeddings = self.int8_index[binary_ids].astype(int)
+
+            scores = query_embedding @ int8_embeddings.T
+
+            indices = (-scores).argsort()[:top_k]
+
+            top_k_indices = binary_ids[indices]
+            top_k_scores = scores[indices]
+
+            return top_k_scores.tolist(), top_k_indices.tolist()
+
+        except Exception as e:
+            print(f"An error occurred while searching semantic similar sentences: {e}")
+
+
+    def _quantize_to_ubinary(
+        self,
+        embeddings: np.ndarray
+    ) -> np.ndarray:
+        """
+        Placeholder private method for ubinary quantization.
+
+        Parameters
+        ----------
+        embeddings : np.ndarray
+            The embeddings to quantize.
+
+        Returns
+        -------
+        np.ndarray
+            The quantized embeddings.
+        """
+        try:
+            ubinary_embeddings = quantize_embeddings(
+                embeddings,
+                "ubinary"
+            )
+            return ubinary_embeddings
+
+        except Exception as e:
+            print(f"An error occurred during ubinary quantization: {e}")
+
+
+    def _quantize_to_int8(
+        self,
+        embeddings: np.ndarray
+    ) -> np.ndarray:
+        """
+        Placeholder private method for int8 quantization.
+
+        Parameters
+        ----------
+        embeddings : np.ndarray
+            The embeddings to quantize.
+
+        Returns
+        -------
+        np.ndarray
+            The quantized embeddings.
+        """
+        try:
+            int8_embeddings = quantize_embeddings(
+                embeddings,
+                "int8"
+            )
+
+            return int8_embeddings
+
+        except Exception as e:
+            print(f"An error occurred during int8 quantization: {e}")
+
+
+    def _save_faiss_index_binary(
+        self,
+        index_path: str
+    ) -> None:
+        """
+        Saves the FAISS binary index to disk.
+
+        This private method is called internally to save the constructed FAISS binary index to the specified file path.
+
+        Parameters
+        ----------
+        index_path : str
+            The path to the file where the binary index should be saved. This value is checked in the public method
+            `create_faiss_index`.
+
+        Returns
+        -------
+        None
+
+        Notes
+        -----
+            This method should not be called directly. It is intended to be used internally by the `create_faiss_index` method.
+        """
+        try:
+            faiss.write_index_binary(
+                self.binary_index,
+                index_path
+            )
+
+            return None
+
+        except Exception as e:
+            print(f"An error occurred during FAISS binary index saving: {e}")
+
+
+    def _save_int8_index(
+        self,
+        index_path: str
+    ) -> None:
+        """
+        Saves the int8_index to disk.
+
+        This private method is called internally to save the constructed int8_index to the specified file path.
+
+        Parameters
+        ----------
+        index_path : str
+            The path to the file where the int8_index should be saved. This value is checked in the public method
+            `_save_int8_index`.
+
+        Returns
+        -------
+        None
+
+        Notes
+        -----
+            This method should not be called directly. It is intended to be used internally by the `_save_int8_index` method.
+        """
+        try:
+            self.int8_index.save(
+                index_path
+            )
+
+            return None
+
+        except Exception as e:
+            print(f"An error occurred during int8_index saving: {e}")

usearch_int8.index

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b756005b791e578b83d0a72d4878ea14cbf8a9cb6b2fd9bb1dede1181d7ae02
+size 145280432