enji-zhou
enji-zhou
test sh: model_name_or_path=/root/qwen_/Qwen1___5-0___5B-Chat/ output_dir=/root/LLMmodels/Qwen1___5-0___5B-Chat-kto-test/ deepspeed --include localhost:3,4,5 --master_port=9909 src/train.py \ --deepspeed /root/llama-efficient-tuning/ds_config_kpo.json \ --stage kto \ --kto_ftx 0.1 \ --model_name_or_path ${model_name_or_path} \ --do_train \ --dataset kto-mix-test \ --template qwen \...
> Hello, could you please handle the package conflicts? Just rename llmtuenr -> llamafactory done
> hmm it seems conflicts still exist, it should be better to open a new PR with the same files 好滴 我重新提交一下
> hmm it seems conflicts still exist, it should be better to open a new PR with the same files new mr:https://github.com/hiyouga/LLaMA-Factory/pull/3785
> 相同配置下，实验下来KTO比DPO更省显存 > > 请问这是为什么？ 我猜是因为DPO是pair输入，KTO是单样本，同样batch size设置下，DPO训练的batch size其实是KTO的两倍
> is it finally fixed? No, it's not fixed yet.
盲猜VENV环境中没装deepspeed