Skip to content

A WIP custom-made retrieval plugin that uses LLaMA and Weaviate with Sentence Transformers as embeddings

License

Notifications You must be signed in to change notification settings

rodobre/llama-weaviate-retrieval

Repository files navigation

What is this ?

A modified version of llama-retrieval-plugin based on Facebook's LLaMA which uses Weaviate and HuggingFace's Sentence Transformers to perform vectorized semantic search.

Required environment variables

export BEARER_TOKEN=$(openssl rand -hex 32)
export WEAVIATE_HOST=http://127.0.0.1
export WEAVIATE_PORT=8080
export WEAVIATE_INDEX=CustomDocument

Tips & Lessons

  • The vector database schema is fully customizable and can be adapted to your needs
  • Different embedding models yield different performances on semantic search (I chose the best performing model for semantic search according to HuggingFace but your mileage may vary)
  • The cloud deployment of Weaviate may work better than the local deploy (ran into some errors that were harder to debug than making the switch to the cloud version)

Contributors & License

About

A WIP custom-made retrieval plugin that uses LLaMA and Weaviate with Sentence Transformers as embeddings

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published