transformers==4.31.0 accelerate tiktoken einops # flash-attention # git clone -b v1.0.8 https://github.com/Dao-AILab/flash-attention # cd flash-attention && pip install . # pip install csrc/layer_norm # pip install csrc/rotary torch==2.0.1 # 2.0.1 safetensors # bitsandbytes # git+https://github.com/TimDettmers/bitsandbytes.git bitsandbytes==0.39.0 transformers_stream_generator scipy loguru about-time