recipe-rag

Find recipe that suits your needs. Built on top of Gemma-2B with RAG hosted on MongoDB. Made using the first 1k recipe from the recipe-nlg dataset.

How to replicate

• The notebook was developed on Google Colab, but can be adapted for use in other environments.
• Create MONGO_URI and HF_TOKEN environment values / secrets for your MongoDB database and Hugging Face account respectively.
• Create a MongoDB collection for hosting the embeddings and assign a vector search index using the guide, while setting path as "embedding", similarity as "cosine" and numDimensions as 1024 to match the embedding model.
• In a GPU environment, run the notebook and follow instructions.

TODO

• Enrich response with the matched recipe.
• Create a simple chat interface.

Reference

• Inspired by a Hugging Face cookbook.