xlang-ai/instructor-embedding

Leverages instruction-following to condition embeddings on task-specific prompts (e.g., "Represent the Science title for retrieval"), enabling a single model to adapt to 70+ diverse tasks without additional finetuning. Built on transformer architecture with three model sizes (base/large/xl) available via Hugging Face, integrated with PyTorch and supporting batch inference with configurable output formats (numpy arrays or tensors). Evaluates across MTEB benchmarks and domain-specific datasets, with support for downstream applications like semantic search, clustering, and text classification through unified embedding API.

2,023 stars. No commits in the last 6 months.