HaoyuWu556

> I changed Makefile:7 as `@pip install --no-deps --verbose --no-cache-dir .` It works for me! it works for me as well, thanks!!

Hi! Is this problem solved? I write a dataset class to inference on my own images. And simply use the same image both for `face_image` and `scene_image`, whose result seems...

我也是这个问题

> > 我也是这个问题 > > batchsize 没满，最后一个drop_last默认丢弃，改为false就好了 Thanks 但是我有两百个样本，batch_size=4,我目前怀疑是bucket和视频精度不匹配的问题。

> @xbyym 数据加载环节有过滤，需要根据自己数据的分布特点来设置bucket_config，代码段： https://github.com/hpcaitech/Open-Sora/blob/main/opensora/datasets/sampler.py#L200-L207 子函数：https://github.com/hpcaitech/Open-Sora/blob/main/opensora/datasets/bucket.py#L74-L120 谢谢~ 请问有什么设置bucket的说明吗 我遇到了一个问题是：视频帧长度不足51的会报错 不知道如何跳过或者设置

Maybe you can also check whether the pretrained model is loaded sucessfully. My training loss of SFT opensora v1.2 start at about 0.4 ~0.5 . But if ckpt not loaded,...

> Thanks for your suggestions! The `num_frames` in my own videos ranges from 2 to 7, which is rather smaller compared to the videos used in Open-Sora. After carefully checking...

Thanks, but I suspect this would be one potential reason of this warning `CUD STATUS NOT SUPPORTED` ? 

Thanks. Then I think the training bugs may have other reasons. [multi gpu training issue](https://github.com/hpcaitech/Open-Sora/issues/565)

同样问题+1 ，自己配的环境有这个问题 但是luchenyun上的镜像没有这个问题