ContextualAI/gritlm

Unifies text embedding and generation in a single model through instruction-based task routing, eliminating the need for separate retrieval and generation models. Built on instruction-tuned language models with pooling mechanisms for dense representations, it supports both zero-shot and instruction-conditioned inference across retrieval-augmented generation, semantic search, and language generation tasks. Available in multiple scales (7B to 8x7B), the framework integrates with Hugging Face's ecosystem and provides caching optimizations for efficient batch processing.

688 stars and 12,353 monthly downloads. Used by 1 other package. No commits in the last 6 months. Available on PyPI.