comfyui

Memory Error

#29
by rsutech - opened

Hi! Have 4090 + 64RAM, loading default ref2va workflow using last ComfyUI on Windows, but every time got Memory Error. Can the problem be like using minimax_h3_ref2va_pruned_fp8_scaled.safetensors insted of int8_convrot ?

ComfyUI Error Report

Error Details

  • Node ID: 620
  • Node Type: UNETLoader
  • Exception Type: MemoryError
  • Exception Message: MemoryError:

Stack Trace

  File "E:\ComfyUI_windows_portable\ComfyUI\execution.py", line 545, in execute
    output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                                                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "E:\ComfyUI_windows_portable\ComfyUI\execution.py", line 344, in get_output_data
    return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "E:\ComfyUI_windows_portable\ComfyUI\execution.py", line 318, in _async_map_node_over_list
    await process_inputs(input_dict, i)

  File "E:\ComfyUI_windows_portable\ComfyUI\execution.py", line 306, in process_inputs
    result = f(**inputs)
             ^^^^^^^^^^^

  File "E:\ComfyUI_windows_portable\ComfyUI\nodes.py", line 988, in load_unet
    model = comfy.sd.load_diffusion_model(unet_path, model_options=model_options)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "E:\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 2210, in load_diffusion_model
    sd, metadata = comfy.utils.load_torch_file(unet_path, return_metadata=True)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

  File "E:\ComfyUI_windows_portable\ComfyUI\comfy\utils.py", line 149, in load_torch_file
    raise e

  File "E:\ComfyUI_windows_portable\ComfyUI\comfy\utils.py", line 129, in load_torch_file
    sd, metadata = load_safetensors(ckpt)
                   ^^^^^^^^^^^^^^^^^^^^^^

  File "E:\ComfyUI_windows_portable\ComfyUI\comfy\utils.py", line 95, in load_safetensors
    header = json.loads(mv[8:8 + header_size].tobytes().decode("utf-8"))
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

System Information

  • ComfyUI Version: 0.30.2
  • Arguments: ComfyUI\main.py --windows-standalone-build --disable-pinned-memory --disable-async-offload --reserve-vram 1
  • OS: win32
  • Python Version: 3.12.10 (tags/v3.12.10:0cc8128, Apr 8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)]
  • Embedded Python: true
  • PyTorch Version: 2.11.0+cu130

Devices

  • Name: cuda:0 NVIDIA GeForce RTX 4090 : cudaMallocAsync
    • Type: cuda
    • VRAM Total: 25756696576
    • VRAM Free: 24159191040
    • Torch VRAM Total: 0
    • Torch VRAM Free: 0
Comfy Org org

I think that's just corrupt model file from failed download or something.

I think that's just corrupt model file from failed download or something.

Thank you for the response, I’ll download it again. I downloaded the fp8_scaled version, deciding that it’s better suited for the 4090. Is that right, or should I use int8_convrot instead?
PS. minimax_h3_fl2va_pruned_int8_convrot works OK

And taking this opportunity I’d also like to personally thank you for your contribution to the development of Comfy — I use your nodes all the time.

Kijay sir, Can we expect a minimax ref2va int4 convrot from official comfy ui (w4a8 version that you are working on!) that would be awesome for low vram and ram users

I think that's just corrupt model file from failed download or something.

Sign up or log in to comment