Yuxuan Zhang

This should be fine, as the 2.4.0 version of PyTorch can also be compiled with CUDA 12.1.

https://github.com/THUDM/CogVideo/blob/main/inference/cli_demo.py follow this and remove the `pipe_image.enable_sequential_cpu_offload()` and use `pipe.to("cuda")`

We only made optimizations on the diffusers version afterwards, as SAT I2V requires too much video memory. In principle, it is the same, both select the first frame of the...

We have supported this feature, allowing the first frame of the video not to be used as the I in I2V for lora trainning. However, this is only code support;...

Do not package it into a zip file; instead, you should directly download this file. Click on the "1" folder and download the tar.gz file.

Exiting now because you already have duplicate files, the correct way to download is 1. Download latest separately 2. Open 1 and download that tar.gz separately. Tsinghua Cloud Disk cannot...

超时还是其他原因，我用的wget能正常下载

单个文件很大，我直接下载文件是在一个网速比较快的环境下载的，大概 20MB/s，没有测试过慢速的时候会不会中断

This error shouldn’t occur because, in the CLI demo, the entire model pipeline is loaded using FP16 (by default), so there shouldn’t be an issue with FP32. Can you print...

This shouldn't be the case. I haven't encountered this situation for the time being because this part of the replacement is reasonable and shouldn't directly cause errors