XTuner大模型单卡低成本微调实战

Finetune简介

XTuner

 

微调训练 

xtuner train ${CONFIG_NAME_OR_PATH}

 

 

部署与测试，将 HuggingFace adapter 合并到大语言模型：

xtuner convert merge ./internlm-chat-7b ./hf ./merged --max-shard-size 2GB
# xtuner convert merge \
#     ${NAME_OR_PATH_TO_LLM} \
#     ${NAME_OR_PATH_TO_ADAPTER} \
#     ${SAVE_PATH} \
#     --max-shard-size 2GB

训练第一轮

第二轮epoch 2

第三轮epoch train3 

跑完训练后，当前路径长这样：

|-- internlm-chat-7b
|-- internlm_chat_7b_qlora_oasst1_e3_copy.py
|-- openassistant-guanaco
|   |-- openassistant_best_replies_eval.jsonl
|   `-- openassistant_best_replies_train.jsonl
`-- work_dirs
    `-- internlm_chat_7b_qlora_oasst1_e3_copy
        |-- 20231101_152923
        |   |-- 20231101_152923.log
        |   `-- vis_data
        |       |-- 20231101_152923.json
        |       |-- config.py
        |       `-- scalars.json
        |-- epoch_1.pth
        |-- epoch_2.pth
        |-- epoch_3.pth
        |-- internlm_chat_7b_qlora_oasst1_e3_copy.py
        `-- last_checkpoint