Xiaoyu Xu

> > 设备型号也请发下，我们可以尝试复现下 > > A800上跑的。onediff是不是会增大内存开销呢，2400*1080分辨率不用onediff能跑，用了onediff后就会出错。 @strint 我们验证下这个分辨率看看 @marigoold 来安排个

@marigoold 可以发下总结

ipa 的 comfyui 支持我们还处于比较早期的状态。当前因为有其它比较紧急的事情，所以暂时 ipa 这里支持不太好。

> 请问nexfort的动态尺寸dynamic还有在支持吗？ 我测了一下 还是会重编译。 Should have no compilation for these new input shape The nexfort backend encounters an exception when dynamically switching resolution to 960x720. 当前是的。现在推荐使用预热的方式，把常用的分辨率都请求一遍，做下预热就会有缓存。

> Nexfort version: none This is not as expected. https://pypi.org/project/nexfort/#history it should be 0.1.dev261

A friend had the same problem, and he fixed this with torch 2.3.0 We will see the torch 2.4.0 problem later(torch 2.4.0 works in my env).

Since pytorch 2.4 is released, nexfort package has been update to pytorch 2.4.0 and cuda 12.1: https://pypi.org/project/nexfort/ @Swarzox @mrrfr @Azru44

https://github.com/siliconflow/onediff/blob/main/onediff_diffusers_extensions/examples/text_to_image_sdxl_reuse_pipe.py#L126-L133 ``` # Update the unet and vae # load_state_dict(state_dict, strict=True, assign=False), assign is False means copying them inplace into the module’s current parameters and buffers. # Reference: https://pytorch.org/docs/stable/generated/torch.nn.Module.html#torch.nn.Module.load_state_dict print("Loading...

> Multi-threaded Python Program Please don't try to use cuda/python/onediff with Multi-thread because it's usually blocked by the Python GIL or not thread-safe. Using multi-process is recommended.

> > > Multi-threaded Python Program > > > > > > Please don't try to use cuda/python/onediff with Multi-thread because it's usually blocked by the Python GIL or not...