hey, is it Q8?

hey, is it Q8?

two ways to solve it:

Option 1: Quick fix in ComfyUI (Keep using your Q8_0 file)

1. Open ComfyUI/custom_nodes/ComfyUI-GGUF/loader.py.
2. Around line 142, find this line:

   if len(shape) <= 1 and tensor.tensor_type == gguf.GGMLQuantizationType.BF16:
   

3. Replace it with:

   if len(shape) <= 1 and tensor.tensor_type not in {gguf.GGMLQuantizationType.F32, gguf.GGMLQuantizationType.F16}:
   

4. Save the file and fully restart ComfyUI (close the command prompt/terminal and launch it again).

Your qwen-image-2.1-Q8_0.gguf will run without any errors.

Option 2: No code editing

If you don't want to edit Python files, you can use qwen-image-2.1-Q6_K.gguf instead. In Q6_K, the norm weights are stored in BF16, so it works completely out of the box with zero modifications, and image generation quality is practically identical.

thanks . first one worket

two ways to solve it:

Lowk W author?