xtuner.md

列出所有内置配置

xtuner list-cfg

微调模型

nohup xtuner train internlm_chat_7b_qlora_self.py --deepspeed deepspeed_zero2 > nohup.out 2>&1 &

PTH模型转换为 HuggingFace 模型

export MKL_SERVICE_FORCE_INTEL=1 export MKL_THREADING_LAYER=GNU xtuner convert pth_to_hf ./internlm_chat_7b_qlora_self.py ./work_dirs/internlm_chat_7b_qlora_self/epoch_1.pth ./hf

模型合并

xtuner convert merge ./internlm-chat-7b ./hf ./merged --max-shard-size 2GB

模型对话

xtuner chat ./merged --prompt-template internlm_chat