Why only your model (other than direct from google) that can run TextGenerated
#6
by komixenon - opened
Basically any other models, gguf and safetensors, have extra 208 tensors. But yours is fine...
These are the error, only error when using Image + Text input. But not so on single text
x = torch.nn.functional.linear(input, weight, bias)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: mat1 and mat2 shapes cannot be multiplied (4096x1152 and 4304x1152)
For gguf it might be missing the vision mmproj file, idk if comfyui gguf supports that yet.
The gemma models here are just the official comfy gemma model quantized using my model converter (which wasn't public yet at the time, but is now)
A version without the vision_modelparts for example will work for text-only tasks, but will error when vision is used. That could be what's going on.