RolePlay-Actor

-模型训练与测试流程

训练文件

>xtuner train /root/personal_assistant/actor/internlm_chat_7b_actor01.py --deepspeed deepspeed_zero2

将训练模型转换为hf格式模型

>>>mkdir hf
export MKL_SERVICE_FORCE_INTEL=1
export MKL_THREADING_LAYER=GNU
xtuner convert pth_to_hf /root/personal_assistant/actor/internlm_chat_7b_actor01.py /root/work_dirs/internlm_chat_7b_actor01/epoch_35.pth /root/personal_assistant/actor01_hfmodel_epoch35

模型文件合并

mkdir /root/personal_assistant/actor01_merge_ep35
xtuner convert merge /root/personal_assistant/model/Shanghai_AI_Laboratory/internlm-chat-7b /root/personal_assistant/actor01_hfmodel_epoch35 /root/personal_assistant/actor01_merge_ep35 --max-shard-size 2GB

终端对话demo测试

xtuner chat /root/personal_assistant/actor01_merge --prompt-template internlm_chat

-模型量化

-主角与配角模型优化

• [*] 模型训练流程与基础demo
• [] 角色模型优化
• [] 接入语音系统