https://github.com/jingyaogong/minimind

预训练:
python train.py
SFT:
python sft_train.py

预训练:
torchrun --nproc_per_node=2 train.py SFT:
torchrun --nproc_per_node=2 sft_train.py

预训练:
deepspeed --include 'localhost:0,1' train.py
SFT:
deepspeed --include 'localhost:0,1' sft_train.py

test_llm.ipynb