rag-wtf/open-text-embeddings

Implements a FastAPI server that wraps HuggingFace sentence-transformer and BGE/E5 models behind an OpenAI-compatible `/embeddings` endpoint, enabling drop-in replacement for OpenAI's embeddings API. Intelligently handles model-specific prefixing strategies—automatically applying query vs. document prefixes based on input type (string vs. list)—critical for optimal performance with instruction-tuned models like BAAI/bge and intfloat/e5 series. Supports both on-premise deployment (CPU/GPU modes) and cloud hosting via AWS Lambda or Modal, with LangChain integration for seamless adoption in RAG pipelines.

166 stars. No commits in the last 6 months.