jdc4429
commented
3 weeks ago For some reason it's now crashing even running SDXL through the API. I have downgraded torch back to 2.3.1 and xformers to .0.0.27 but it's still crashing. It always seems to crash after Requested to load AutoencoderKL.
Error: model_type EPS Using xformers attention in VAE Using xformers attention in VAE Requested to load SDXLClipModel Loading 1 new model Requested to load SDXL Loading 1 new model 100%|█████████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:21<00:00, 2.71s/it] Requested to load AutoencoderKL Loading 1 new model [START] Security scan [DONE] Security scan
Expected Behavior
Image created
Actual Behavior
Error:
model weight dtype torch.float8_e4m3fn, manual cast: torch.float16 modeltype FLOW Using xformers attention in VAE Using xformers attention in VAE Requested to load FluxClipModel Loading 1 new model Requested to load Flux Loading 1 new model 100%|██████████████████████████████████████████████████████████████████████████████| 6/6 [01:32<00:00, 15.39s/it] Using xformers attention in VAE Using xformers attention in VAE Requested to load AutoencoderKL Loading 1 new model !!! Exception during processing!!! Given groups=1, weight of size [4, 4, 1, 1], expected input[1, 16, 160, 90] to have 4 channels, but got 16 channels instead Traceback (most recent call last): File "C:\inetpub\wwwroot\ComfyUI\ComfyUI\execution.py", line 152, in recursive_execute output_data, output_ui = get_output_data(obj, input_data_all) File "C:\inetpub\wwwroot\ComfyUI\ComfyUI\execution.py", line 82, in get_output_data return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True) File "C:\inetpub\wwwroot\ComfyUI\ComfyUI\execution.py", line 75, in map_node_over_list results.append(getattr(obj, func)(slice_dict(input_data_all, i))) File "C:\inetpub\wwwroot\ComfyUI\ComfyUI\nodes.py", line 270, in decode return (vae.decode(samples["samples"]), ) File "C:\inetpub\wwwroot\ComfyUI\ComfyUI\comfy\sd.py", line 322, in decode pixel_samples[x:x+batch_number] = self.process_output(self.first_stage_model.decode(samples).to(self.output_device).float()) File "C:\inetpub\wwwroot\ComfyUI\ComfyUI\comfy\ldm\models\autoencoder.py", line 199, in decode dec = self.post_quant_conv(z) File "C:\inetpub\wwwroot\ComfyUI\python_embeded\lib\site-packages\torch\nn\modules\module.py", line 1553, in _wrapped_call_impl return self._call_impl(*args, *kwargs) File "C:\inetpub\wwwroot\ComfyUI\python_embeded\lib\site-packages\torch\nn\modules\module.py", line 1562, in _call_impl return forward_call(args, kwargs) File "C:\inetpub\wwwroot\ComfyUI\ComfyUI\comfy\ops.py", line 93, in forward return super().forward(*args, **kwargs) File "C:\inetpub\wwwroot\ComfyUI\python_embeded\lib\site-packages\torch\nn\modules\conv.py", line 458, in forward return self._conv_forward(input, self.weight, self.bias) File "C:\inetpub\wwwroot\ComfyUI\python_embeded\lib\site-packages\torch\nn\modules\conv.py", line 454, in _conv_forward return F.conv2d(input, weight, bias, self.stride, RuntimeError: Given groups=1, weight of size [4, 4, 1, 1], expected input[1, 16, 160, 90] to have 4 channels, but got 16 channels instead
Prompt executed in 194.95 seconds
Steps to Reproduce
Run workflow... Flux full checkpoint API test.json
Debug Logs
Other
To confirm, I can run this workflow myself in the web interface without any issues. But when I copy the workflow to an API I get the above error. Some how the vae does not appear to be matching I believe.. 16 channels instead of 4.. I don't understand how this works in the web interface but not through the API.