HomeProjectsArticlesPK BoardAbout
Explore Projects

FlagEmbedding

Active
GitHub Python MIT

Description

Open-source BGE series embedding models and retrieval tools from BAAI, providing state-of-the-art text embeddings and rerankers for Chinese and English, widely used in RAG systems and agent retrieval pipelines.

Key Features

  • State-of-the-art multilingual embeddings (BGE-M3) supporting 100+ languages with dense, lexical, and multi-vector retrieval
  • BGE-VL multimodal embeddings for text-to-image, image-to-text, and cross-modal visual search
  • Lightweight rerankers with token compression and layerwise operations for efficient resource usage
  • In-context learning capable embeddings (bge-en-icl) that encode semantically richer queries
  • Comprehensive toolkit covering embedding, retrieval, reranking, and long-context LLM extensions
  • Fully open-source under MIT license for both academic and commercial use

Use Cases

💡 Building RAG pipelines with high-quality text retrieval for enterprise knowledge bases
💡 Multilingual semantic search across documents in 100+ languages
💡 Visual search applications using BGE-VL for image-text cross-modal retrieval
💡 Long-context LLM augmentation via Activation-Beacon context extension
💡 Search engine result reranking for improved relevance and precision

Strengths & Limitations

✅ Strengths

  • • Actively maintained, recent updates
  • • High community interest (12.1k stars)
  • • Permissive open-source license (MIT)
  • • Established track record (3 years in production)

⚠️ Limitations

  • • High issue backlog (907 open issues)

Tags

embeddings bge retrieval reranker rag sentence-embeddings

Categories

📚 RAG Tools

Quick Start

pip install FlagEmbedding && python -c "from FlagEmbedding import BGEM3FlagModel; model = BGEM3FlagModel('BAAI/bge-m3'); output = model.encode(['Hello world']); print(output['dense_vecs'].shape)"
Visit GitHub Visit Website

Related Projects

LanceDB

11.2k · Rust
Active A

An open-source embedded retrieval library for multimodal AI with zero server configuration, using the Lance columnar format for efficient vector search and filtering, ideal for agent memory and RAG applications.

vector-databaseembeddingsretrieval +2
  • · Millisecond vector search with state-of-the-art indexing for billions of vectors
  • · Comprehensive search: vector similarity, full-text search, and SQL queries in one platform
  • · Multimodal support — store and query text, images, videos, point clouds, and more
View full details → View on GitHub

FastEmbed

3.1k · Python
Active A

Lightweight, fast Python embedding library from the Qdrant team. Built on ONNX Runtime, requires no GPU and avoids GBs of PyTorch dependencies, making it ideal for serverless runtimes like AWS Lambda.

embeddingsonnxpython +3
  • · Lightweight — minimal external dependencies, no PyTorch, no GPU required — ideal for serverless runtimes like AWS Lambda
  • · Fast — built on ONNX Runtime, faster than PyTorch
  • · Accurate — default model outperforms OpenAI Ada-002 on the MTEB leaderboard
View on GitHub

LightRAG

38.9k · Python
Active A+

LightRAG is a simple and fast Retrieval-Augmented Generation framework using graph-enhanced retrieval, published at EMNLP 2025.

raggraphretrieval +2
  • · Graph-enhanced retrieval — Dual-level retrieval (local/global) using knowledge graph entities and relationships, more precise than vector search
  • · Four text chunking strategies — Fixed, Recursive, Vector, and Paragraph chunking adapted to different document types
  • · Multi-backend storage — Neo4j, PostgreSQL, MongoDB, OpenSearch, JSON KV Store and other storage backends supported
View full details → View on GitHub

RAG Techniques

29.1k · Jupyter Notebook
Active A+

NirDiamant's RAG_Techniques repository systematically collects and demonstrates cutting-edge retrieval-augmented generation techniques and engineering practices.

ragretrievaltechniques +2
  • · Systematic taxonomy — techniques organised by simple / advanced / modular levels for targeted learning
  • · Runnable notebooks — every technique ships with a Jupyter notebook and minimal runnable code
  • · Evaluation & comparison — retrieval quality and answer quality evaluation scripts with visualisation
View full details → View on GitHub

Related Articles

RAG重排序向量检索

Reranking and Hybrid Search: A Practical Guide to Production RAG Retrieval

A production-engineering deep dive into RAG two-stage retrieval: Reranker model selection, Hybrid Search fusion strategies, Qdrant sparse+dense implementation, and offline evaluation set design -- taking production RAG recall from 60% to 90%.

Read more →