Christopher
Christopher
Thanks @aredden for your explanation, but look, first I'm setting all seeds for all generators (random, torch, etc) as I'm usually doing for diffusers. Then I run a first inference,...
I found. This is due to fast acc. Thx!
I have reproductibility now on H100 but still not on H200, that's why I'm opening again the issue.
Sounds like h100 and h200 have slightly different architectures and implementations ?
Sorry I'm not sure to understand the question if it's for me. I'm running inferences with quantized flux schnell model and I have different expected images when scaled mm with...
hi @orionlee your PR does not work. Sometimes, pos is None, which leads to an error.
Same The error appears during loading `self.texgen_pipeline = Hunyuan3DPaintPipeline.from_pretrained('tencent/Hunyuan3D-2')` Why turbo model while I'm loading non turbo model ?
Same problem for me
Doesn't work for me. Is the error linked to the warning: ``` W1029 20:55:47.576000 79341 torch/utils/cpp_extension.py:531] There are no /usr/bin/g++-14 version bounds defined for CUDA version 13.0 building 'flash_attn_3._C' extension...