TanGentleman / Augmenta

Automate RAG-powered workflows
MIT License
1 stars 0 forks source link

Use dynamic m2-bert model based on chunk size #38

Open TanGentleman opened 6 months ago

TanGentleman commented 6 months ago

For tasks like these, the m2-bert-80M-8k-retrieval, as well as 2k or 32k versions seem to perform pretty well. I can choose the model in the RagSettings class after the chunk size has been established. If chunk size is changed from anywhere, it should automatically update the retrieval model appropriately. This makes use of the decorator functions, OOP, and the efficient manifest checks/caching.