torch==2.6.0 torchvision==0.21.0 transformers==4.52.4 accelerate deepspeed datasets evaluate peft safetensors flashinfer-python>=0.2.12 xformers triton>=3.0.0 tqdm numpy scipy pandas scikit-learn sentencepiece huggingface-hub omegaconf liger_kernel termcolor nest_asyncio ipykernel math_verify wandb tensorboard