torch.OutOfMemoryError: Allocation on device

[bluesky61]

My PC uses GTX3070.

When I start running the GGUF workflow, I get an OutOfMemoryError at the Negative Prompt, but what's even weirder is that if I press the "Queue" button again, it goes through and generates the image just fine.

This happens at least 5 times out of 10, and I think the owner need to find out why this is happening.

ComfyUI Error Report

Error Details

• Node ID: 33
• Node Type: CLIPTextEncode
• Exception Type: torch.OutOfMemoryError
• Exception Message: Allocation on device

Stack Trace

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\execution.py", line 327, in execute
    output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)
                                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\execution.py", line 202, in get_output_data
    return_values = _map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\execution.py", line 174, in _map_node_over_list
    process_inputs(input_dict, i)

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\execution.py", line 163, in process_inputs
    results.append(getattr(obj, func)(**inputs))
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\nodes.py", line 67, in encode
    return (clip.encode_from_tokens_scheduled(tokens), )
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 146, in encode_from_tokens_scheduled
    pooled_dict = self.encode_from_tokens(tokens, return_pooled=return_pooled, return_dict=True)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 207, in encode_from_tokens
    self.load_model()

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 240, in load_model
    model_management.load_model_gpu(self.patcher)

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 555, in load_model_gpu
    return load_models_gpu([model])
           ^^^^^^^^^^^^^^^^^^^^^^^^

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 550, in load_models_gpu
    loaded_model.model_load(lowvram_model_memory, force_patch_weights=force_patch_weights)

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 366, in model_load
    self.model_use_more_vram(use_more_vram, force_patch_weights=force_patch_weights)

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 395, in model_use_more_vram
    return self.model.partially_load(self.device, extra_memory, force_patch_weights=force_patch_weights)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\comfy\model_patcher.py", line 759, in partially_load
    raise e

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\comfy\model_patcher.py", line 756, in partially_load
    self.load(device_to, lowvram_model_memory=current_used + extra_memory, force_patch_weights=force_patch_weights, full_load=full_load)

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-GGUF\nodes.py", line 84, in load
    super().load(*args, force_patch_weights=True, **kwargs)

  File "H:\ComfyUI\ComfyUI_windows_portable\ComfyUI\comfy\model_patcher.py", line 607, in load
    x[2].to(device_to)

  File "H:\ComfyUI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1340, in to
    return self._apply(convert)
           ^^^^^^^^^^^^^^^^^^^^

  File "H:\ComfyUI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 927, in _apply
    param_applied = fn(param)

[al-swaiti]

i think now its better to use gguf diffusion model loader with native comfyui clip loader (fp8 or fp16)

[bluesky61]

i think now its better to use gguf diffusion model loader with native comfyui clip loader (fp8 or fp16)

The following CLIP Loader exists in Comfy Core.

• Load CLIP
• DUALCLIPLoader
• TripleCLIPLoader

However, none of them can select the GGUF (t5-v1_1-xxl-encoder-f16.gguf) model.

I would appreciate it if you could tell me which node you are referring to.

[al-swaiti]

gguf clip models not supported in new comfyui versions

[bluesky61]

gguf clip models not supported in new comfyui versions

I don't understand, I just tried to create an image using the DUALCLIPLoader(GGUF) node from ComfyUI-GGUF a while ago and it worked fine.
I am using ComfyUI v0.3.14, the latest version.

[al-swaiti]

try this trick to pypass negative prompt or using this workflow where NEG not needed (drop image to comfyui)https://comfyanonymous.github.io/ComfyUI_examples/flux/flux_dev_example.png

[bluesky61]

Thank you very much. With the {ConditioningZeroOut], the excution time is reduced. Although the "OutOfMemoryError" appears at Positive prompt this time. :)