Kingsley
Kingsley
you can refer to https://github.com/hiyouga/LLaMA-Factory/issues/8365#issuecomment-3251631509
It is not recommend to use llamafactory to do ppo with moe-type model.
> 或者多卡并行有什么配置可以让token也切成四分分配到四个gpu上呢 Sequence parallelism will be supported in the next version.
I reproduced your exp with following config without catching this issue. `hardware env: 2V100` ```yaml ### model model_name_or_path: ./Qwen2.5-Omni-7B image_max_pixels: 262144 video_max_pixels: 16384 trust_remote_code: true ### method stage: sft do_train:...
I am unfamiliar with MLLM role-playing, but I guess it is inconsistent with model training because these mllm models use a fixed textual system prompt (see their chat templates). Can...
Thanks for your explanation. I will check it later. I just added some prefix images in user column and restart this experiment without catching this. I think it should be...
> An alternative way is that, we put this role instruction on the first `user` message of a dialogue, and `assistant` reply "OK". And then the real chat begins from...
> Thank you for your experiment. > But I still want to know how to achieve mmdata in sys_msg when training.😂 Is it too hard? If it's too hard, then...
> I'm now converting my videos into vision_only_videos following [#7638](https://github.com/hiyouga/LLaMA-Factory/pull/7638). And I guess the hanging problem with my own dataset is mainly caused by this. Hope I will come back...