alex13by
alex13by
It may be possible to support some special workflows in the form of lists, such as SDXL, LCM, cascade, etc.
Now my flux is completely unusable for forge. Request guidance After I tried to move clip to vae: Model selected: {'checkpoint_info': {'filename': 'D:\\AI\\webui_forge\\webui\\models\\Stable-diffusion\\flux\\flux1-dev-fp8.safetensors', 'hash': '26acbda5'}, 'additional_modules': ['D:\\AI\\webui_forge\\webui\\models\\VAE\\clip\\clip_l.safetensors', 'D:\\AI\\webui_forge\\webui\\models\\VAE\\clip\\t5xxl_fp8_e4m3fn.safetensors', 'D:\\AI\\webui_forge\\webui\\models\\VAE\\sd3VAE_v10.safetensors'], 'unet_storage_dtype':...
> 更新并重试 nf4可以正常运行，fp8直接崩溃退出： Model selected: {'checkpoint_info': {'filename': 'D:\\AI\\webui_forge\\webui\\models\\Stable-diffusion\\flux\\flux1-dev-fp8.safetensors', 'hash': '26acbda5'}, 'additional_modules': ['D:\\AI\\webui_forge\\webui\\models\\VAE\\clip\\t5xxl_fp16.safetensors', 'D:\\AI\\webui_forge\\webui\\models\\VAE\\clip\\clip_l.safetensors', 'D:\\AI\\webui_forge\\webui\\models\\VAE\\sd3VAE_v10.safetensors'], 'unet_storage_dtype': torch.float8_e4m3fn} Loading Model: {'checkpoint_info': {'filename': 'D:\\AI\\webui_forge\\webui\\models\\Stable-diffusion\\flux\\flux1-dev-fp8.safetensors', 'hash': '26acbda5'}, 'additional_modules': ['D:\\AI\\webui_forge\\webui\\models\\VAE\\clip\\t5xxl_fp16.safetensors', 'D:\\AI\\webui_forge\\webui\\models\\VAE\\clip\\clip_l.safetensors', 'D:\\AI\\webui_forge\\webui\\models\\VAE\\sd3VAE_v10.safetensors'], 'unet_storage_dtype': torch.float8_e4m3fn} StateDict...
看到又有新的更新了，fp8开始推理没有报错了，VAE / Text Encoder中选取的clip和vae，Euler20步，但出来的图像模糊就像采样器不支持一样，换nf4完全正常，
There is no way to search in the translated language, and the display is also inconvenient
> Yes, it's supported. You need to install the specified version of Torch by yourself. I've tested it on the latest version of Torch. The repo uses a low version...
> 将 pytorch 降级到 2.3.1 并重试。 Same problem under torch2.31 cu121
> 将 pytorch 降级到 2.3.1 并重试。 [comfyui.log](https://github.com/user-attachments/files/16808654/comfyui.log) This is the full log of the 2.4.0 run, including inference prompts, Before upgrading to 2.4.0 is torch 2.3.1+CU121
The most simplified workflow and CLIP selection is normal, if changed to 22G DEV original version using fp8 will not black plot, but using 11G fp8 version will black plot,...
Can you give a complete configuration example? The purpose of environment variables is not fully understood