BM25S Index

This is a BM25S index created with the bm25s library (version 0.2.3), an ultra-fast implementation of BM25. It can be used for lexical retrieval tasks.

BM25S Related Links:

Installation

You can install the bm25s library with pip:

pip install "bm25s==0.2.3"

# Include extra dependencies like stemmer
pip install "bm25s[full]==0.2.3"

# For huggingface hub usage
pip install huggingface_hub

Loading a `bm25s` index

You can use this index for information retrieval tasks. Here is an example:

import bm25s
from bm25s.hf import BM25HF

# Load the index
retriever = BM25HF.load_from_hub("ylkhayat/bm25s-caselaw-us-and-veterans")

# You can retrieve now
query = "a cat is a feline"
results = retriever.retrieve(bm25s.tokenize(query), k=3)

Saving a `bm25s` index

You can save a bm25s index to the Hugging Face Hub. Here is an example:

import bm25s
from bm25s.hf import BM25HF

corpus = [
    "a cat is a feline and likes to purr",
    "a dog is the human's best friend and loves to play",
    "a bird is a beautiful animal that can fly",
    "a fish is a creature that lives in water and swims",
]

retriever = BM25HF(corpus=corpus)
retriever.index(bm25s.tokenize(corpus))

token = None  # You can get a token from the Hugging Face website
retriever.save_to_hub("ylkhayat/bm25s-caselaw-us-and-veterans", token=token)

Advanced usage

You can leverage more advanced features of the BM25S library during load_from_hub:

# Load corpus and index in memory-map (mmap=True) to reduce memory
retriever = BM25HF.load_from_hub("ylkhayat/bm25s-caselaw-us-and-veterans", load_corpus=True, mmap=True)

# Load a different branch/revision
retriever = BM25HF.load_from_hub("ylkhayat/bm25s-caselaw-us-and-veterans", revision="main")

# Change directory where the local files should be downloaded
retriever = BM25HF.load_from_hub("ylkhayat/bm25s-caselaw-us-and-veterans", local_dir="/path/to/dir")

# Load private repositories with a token:
retriever = BM25HF.load_from_hub("ylkhayat/bm25s-caselaw-us-and-veterans", token=token)

Tokenizer

If you have saved a Tokenizer object with the index using the following approach:

from bm25s.hf import TokenizerHF

token = "your_hugging_face_token"
tokenizer = TokenizerHF(corpus=corpus, stopwords="english")
tokenizer.save_to_hub("ylkhayat/bm25s-caselaw-us-and-veterans", token=token)

# and stopwords too
tokenizer.save_stopwords_to_hub("ylkhayat/bm25s-caselaw-us-and-veterans", token=token)

Then, you can load the tokenizer using the following code:

from bm25s.hf import TokenizerHF

tokenizer = TokenizerHF(corpus=corpus, stopwords=[])
tokenizer.load_vocab_from_hub("ylkhayat/bm25s-caselaw-us-and-veterans", token=token)
tokenizer.load_stopwords_from_hub("ylkhayat/bm25s-caselaw-us-and-veterans", token=token)

Stats

This dataset was created using the following data:

Statistic	Value
Number of documents	366752
Number of tokens	57736416
Average tokens per document	157.43

Parameters

The index was created with the following parameters:

Parameter	Value
k1	`1.5`
b	`0.75`
delta	`0.5`
method	`lucene`
idf method	`lucene`

Citation

To cite bm25s, please use the following bibtex:

@misc{lu_2024_bm25s,
      title={BM25S: Orders of magnitude faster lexical search via eager sparse scoring}, 
      author={Xing Han Lù},
      year={2024},
      eprint={2407.03618},
      archivePrefix={arXiv},
      primaryClass={cs.IR},
      url={https://arxiv.org/abs/2407.03618}, 
}