Skip to content

Latest commit

 

History

History
12 lines (10 loc) · 530 Bytes

README.md

File metadata and controls

12 lines (10 loc) · 530 Bytes

Haku-Phi: A Phi playground

Haku - 珀

Original model: https://huggingface.co/microsoft/phi-1_5

What is this repo

Just a playground to place lot of things about phi. Maybe trainer/inference ui/optimizations/some addon archs...etc

Plans:

  • optimizations: like xformers/pytorch sdp attn
  • trainer: trainer for LLM/VQA/Image Captioner
  • inference ui: Like what I have done on Guanaco-7B (not the Q-LoRA one).
  • some addon archs: For VQA or t2i model.