torchvision==0.20.0 transformers==4.51.1 accelerate sentencepiece attrdict einops tiktoken blobfile https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.5cxx11abiFALSE-cp310-cp310-linux_x86_64.whl # for gradio demo gradio gradio-client mdtex2html pypinyin tqdm colorama Pygments markdown SentencePiece