FlagEmbedding README

The project's README from github.com/FlagOpen/FlagEmbedding. As one markdown file: README.md.

⚡️BGE: One-Stop Retrieval Toolkit For Search and RAG

bge_logo

Build Build License Build Build

News | Installation | Quick Start | Community | Projects | Model List | Contributor | Citation | License

English | 中文

News

More
  • 6/7/2024: Release a new benchmark MLVU, the first comprehensive benchmark specifically designed for long video understanding. MLVU features an extensive range of video durations, a diverse collection of video sources, and a set of evaluation tasks uniquely tailored for long-form video understanding. :fire:
  • 5/21/2024: Release a new benchmark AIR-Bench together with Jina AI, Zilliz, HuggingFace, and other partners. AIR-Bench focuses on a fair out-of-distribution evaluation for Neural IR & RAG. It generates the synthetic data for benchmarking w.r.t. diverse domains and languages. It is dynamic and will be updated on regular basis. Leaderboard :fire:
  • 4/30/2024: Release Llama-3-8B-Instruct-80K-QLoRA, extending the context length of Llama-3-8B-Instruct from 8K to 80K via QLoRA training on a few synthesized long-context data. The model achieves remarkable performance on various long-context benchmarks. Code :fire:
  • 3/18/2024: Release new rerankers, built upon powerful M3 and LLM (GEMMA and MiniCPM, not so large actually :smiley:) backbones, supporitng multi-lingual processing and larger inputs, massive improvements of ranking performances on BEIR, C-MTEB/Retrieval, MIRACL, LlamaIndex Evaluation :fire:
  • 3/18/2024: Release Visualized-BGE, equipping BGE with visual capabilities. Visualized-BGE can be utilized to generate embeddings for hybrid image-text data. :fire:
  • 1/30/2024: Release BGE-M3, a new member to BGE model series! M3 stands for Multi-linguality (100+ languages), Multi-granularities (input length up to 8192), Multi-Functionality (unification of dense, lexical, multi-vec/colbert retrieval). It is the first embedding model which supports all three retrieval methods, achieving new SOTA on multi-lingual (MIRACL) and cross-lingual (MKQA) benchmarks. Technical Report and Code. :fire:
  • 1/9/2024: Release Activation-Beacon, an effective, efficient, compatible, and low-cost (training) method to extend the context length of LLM. Technical Report
  • 12/24/2023: Release LLaRA, a LLaMA-7B based dense retriever, leading to state-of-the-art performances on MS MARCO and BEIR. Model and code will be open-sourced. Please stay tuned. Technical Report and Code
  • 11/23/2023: Release LM-Cocktail, a method to maintain general capabilities during fine-tuning by merging multiple language models. Technical Report
  • 10/12/2023: Release LLM-Embedder, a unified embedding model to support diverse retrieval augmentation needs for LLMs. Technical Report
  • 09/15/2023: The technical report of BGE has been released
  • 09/15/2023: The massive training data of BGE has been released
  • 09/12/2023: New models:
    • New reranker model: release cross-encoder models BAAI/bge-reranker-base and BAAI/bge-reranker-large, which are more powerful than embedding model. We recommend to use/fine-tune them to re-rank top-k documents returned by embedding models.
    • update embedding model: release bge-*-v1.5 embedding model to alleviate the issue of the similarity distribution, and enhance its retrieval ability without instruction.
  • 09/07/2023: Update fine-tune code: Add script to mine hard negatives and support adding instruction during fine-tuning.
  • 08/09/2023: BGE Models are integrated into Langchain, you can use it like this; C-MTEB leaderboard is available.
  • 08/05/2023: Release base-scale and small-scale models, best performance among the models of the same size 🤗
  • 08/02/2023: Release bge-large-*(short for BAAI General Embedding) Models, rank 1st on MTEB and C-MTEB benchmark! :tada: :tada:
  • 08/01/2023: We release the Chinese Massive Text Embedding Benchmark (C-MTEB), consisting of 31 test dataset.

BGE (BAAI General Embedding) focuses on retrieval-augmented LLMs, consisting of the following projects currently:

projects

Installation

Using pip:

If you do not want to finetune the models, you can install the package without the finetune dependency:

pip install -U FlagEmbedding

If you want to finetune the models, you can install the package with the finetune dependency:

pip install -U FlagEmbedding[finetune]

Install from sources:

Clone the repository and install

git clone https://github.com/FlagOpen/FlagEmbedding.git
cd FlagEmbedding
# If you do not need to finetune the models, you can install the package without the finetune dependency:
pip install  .
# If you want to finetune the models, install the package with the finetune dependency:
# pip install  .[finetune]

For development in editable mode:

# If you do not need to finetune the models, you can install the package without the finetune dependency:
pip install -e .
# If you want to finetune the models, install the package with the finetune dependency:
# pip install -e .[finetune]

Quick Start

First, load one of the BGE embedding model:

from FlagEmbedding import FlagAutoModel

model = FlagAutoModel.from_finetuned('BAAI/bge-base-en-v1.5',
                                      query_instruction_for_retrieval="Represent this sentence for searching relevant passages:",
                                      use_fp16=True)

Then, feed some sentences to the model and get their embeddings:

sentences_1 = ["I love NLP", "I love machine learning"]
sentences_2 = ["I love BGE", "I love text retrieval"]
embeddings_1 = model.encode(sentences_1)
embeddings_2 = model.encode(sentences_2)

Once we get the embeddings, we can compute similarity by inner product:

similarity = embeddings_1 @ embeddings_2.T
print(similarity)

For more details, you can refer to embedder inference, reranker inference, embedder finetune, reranker fintune, evaluation.

If you're unfamiliar with any of related concepts, please check out the tutorial. If it's not there, let us know.

For more interesting topics related to BGE, take a look at research.

Community

We are actively maintaining the community of BGE and FlagEmbedding. Let us know if you have any suggessions or ideas!

Currently we are updating the tutorials, we aim to create a comprehensive and detailed tutorial for beginners on text retrieval and RAG. Stay tuned!

The following contents are releasing in the upcoming weeks:

The whole tutorial roadmap

Model List

bge is short for BAAI general embedding.

Model Language Description query instruction for retrieval
BAAI/bge-en-icl English A LLM-based embedding model with in-context learning capabilities, which can fully leverage the model's potential based on a few shot examples Provide instructions and few-shot examples freely based on the given task.
BAAI/bge-multilingual-gemma2 Multilingual A LLM-based multilingual embedding model, trained on a diverse range of languages and tasks. Provide instructions based on the given task.
BAAI/bge-m3 Multilingual Multi-Functionality(dense retrieval, sparse retrieval, multi-vector(colbert)), Multi-Linguality, and Multi-Granularity(8192 tokens)
LM-Cocktail English fine-tuned models (Llama and BGE) which can be used to reproduce the results of LM-Cocktail
BAAI/llm-embedder English a unified embedding model to support diverse retrieval augmentation needs for LLMs See README
BAAI/bge-reranker-v2-m3 Multilingual a lightweight cross-encoder model, possesses strong multilingual capabilities, easy to deploy, with fast inference.
BAAI/bge-reranker-v2-gemma Multilingual a cross-encoder model which is suitable for multilingual contexts, performs well in both English proficiency and multilingual capabilities.
BAAI/bge-reranker-v2-minicpm-layerwise Multilingual a cross-encoder model which is suitable for multilingual contexts, performs well in both English and Chinese proficiency, allows freedom to select layers for output, facilitating accelerated inference.
BAAI/bge-reranker-v2.5-gemma2-lightweight Multilingual a cross-encoder model which is suitable for multilingual contexts, performs well in both English and Chinese proficiency, allows freedom to select layers, compress ratio and compress layers for output, facilitating accelerated inference.
BAAI/bge-reranker-large Chinese and English a cross-encoder model which is more accurate but less efficient
BAAI/bge-reranker-base Chinese and English a cross-encoder model which is more accurate but less efficient
BAAI/bge-large-en-v1.5 English version 1.5 with more reasonable similarity distribution Represent this sentence for searching relevant passages:
BAAI/bge-base-en-v1.5 English version 1.5 with more reasonable similarity distribution Represent this sentence for searching relevant passages:
BAAI/bge-small-en-v1.5 English version 1.5 with more reasonable similarity distribution Represent this sentence for searching relevant passages:
BAAI/bge-large-zh-v1.5 Chinese version 1.5 with more reasonable similarity distribution 为这个句子生成表示以用于检索相关文章：
BAAI/bge-base-zh-v1.5 Chinese version 1.5 with more reasonable similarity distribution 为这个句子生成表示以用于检索相关文章：
BAAI/bge-small-zh-v1.5 Chinese version 1.5 with more reasonable similarity distribution 为这个句子生成表示以用于检索相关文章：
BAAI/bge-large-en English Embedding Model which map text into vector Represent this sentence for searching relevant passages:
BAAI/bge-base-en English a base-scale model but with similar ability to bge-large-en Represent this sentence for searching relevant passages:
BAAI/bge-small-en English a small-scale model but with competitive performance Represent this sentence for searching relevant passages:
BAAI/bge-large-zh Chines

FlagEmbedding (FlagOpen/FlagEmbedding)

Searching Database…

12.3K ·GitHub
FlagEmbedding
What is this?FlagEmbedding (BGE) is a one-stop toolkit from BAAI for retrieval and retrieval-augmented LLMs, providing embedding models, rerankers, finetuning scripts, and evaluation tools for search and RAG.
Is this a model collection, a Python library, or both?It is both: a Python package (pip-installable) that wraps inference/finetuning code, plus a family of pretrained embedding and reranker models hosted on Hugging Face (BAAI org).
How do I get started with it?Install via pip (with or without the finetune extra), or clone the repo and install from source in editable mode, then load a BGE embedding model and feed it sentences to get embeddings.
Do I need anything beyond Python to use it?Just Python and pip; models are pulled from Hugging Face (BAAI repos), and an optional 'finetune' extra is needed only if you plan to fine-tune your own embedder or reranker.
Which model should I actually pick from the list?There are many overlapping model variants (v1, v1.5, M3, ICL, gemma2-based) with different languages and tradeoffs (accuracy vs. speed), so picking the right one takes reading the model table rather than being obvious.
Does it support languages other than English and Chinese?Yes, several models like BGE-M3 and bge-multilingual-gemma2 are explicitly multilingual, supporting 100+ languages alongside the English- and Chinese-specific v1.5 models.
What does BGE-M3 add over the regular embedding models?BGE-M3 unifies three retrieval methods (dense, sparse/lexical, and multi-vector ColBERT-style) in one model, supports 100+ languages, and handles inputs up to 8192 tokens.
What is the reranker used for versus the embedder?The reranker is a cross-encoder that re-scores the top-k documents already retrieved by an embedding model, typically giving more accurate but slower results than embedding-based similarity alone.
Can I use BGE with LangChain?Yes, the README notes BGE models are integrated into LangChain so they can be used directly as embedding providers in that framework.
What is Visualized-BGE?Visualized-BGE equips BGE with visual capabilities, letting it generate embeddings for hybrid image-text data for multimodal search.
What are similar projects?txtai, LLMSearchRecommender, fastembed-rs, sentence-transformers, SkillCorpus, memvid, PageIndex, MSA
Which topics is it in?information-retrieval, retrieval-augmented-generation, vector-search, semantic-search, embeddings, fine-tuning, transformer, multilingual, nlp, rag, huggingface, benchmark
Creator
FlagOpen
Size
52.2 MB
Stars
12.3K
Updated
CommitAug 24Releasev1.4.2Aug 24
Age
3y
Commits · 90 Days