FlashRAG icon indicating copy to clipboard operation
FlashRAG copied to clipboard

[Update] new functions and bug fixes

Open yeahjack opened this issue 4 months ago • 0 comments

During this PR, I made the following changes:

  1. Add support for flash-attention-2 in HFCausalLMGenerator, and add Llama-3 special token when initializing the model.
  2. Add refiner on multi GPU, hence accelerating the refining process by a lot.
  3. Fix a bug for selective-context for list out of bounds.
  4. Add paths for models in recomp and fix loading.
  5. Fix a bug for intermediate_data.json saving when using bm25s.

yeahjack avatar Oct 06 '24 17:10 yeahjack