Zhenyu Wang (Luffy)
Zhenyu Wang (Luffy)
I just downgrade `deepspeed==0.16.4` by chance and it seems to be able to run finally. I'm confirming it again. Is there any difference between these two versions that cause this...
I tested `deepspeed==0.15.0` and it can also run. It seems it's the prob with ds=0.16.5. And I'm sorry about the `ds_report` info above because I ran it after I downgrade...
This issue has been solved by downgrading deepspeed version from 0.16.5 to 0.16.4 or 0.15.0. Would not close this issue for waiting for deepspeed team to handle this bug. Once...
> [@Luffy-ZY-Wang](https://github.com/Luffy-ZY-Wang), thank for reporting this issue. I am concerned by the effort to accurately fill in the placeholders in your repro. Do you know if this issue can be...
I can reproduce this issue steadily with the following config: ```yaml ### model model_name_or_path: /148Dataset/data-wang.zhenyu/models/ModelScope_download/Qwen2.5-VL-7B-Instruct image_max_pixels: 16384 video_max_pixels: 16384 trust_remote_code: true video_fps: 2.0 # won't matter if comment out video_maxlen:...
Any progress yet? I'm still facing this with ds=0.16.7
Looking forward to the update!
> 已支持 @hiyouga 大佬能发一下是哪个版本支持了吗，现在0.9.3.dev0好像会有这个`AssertionError: Qwen2AudioForConditionalGeneration does not support LoRA yet.`报错
sft with only 2*4090
Hi, @Kuangdd01 Could you please do me a favor by sparing some time to watch this?