Please Help

[balkerz666]

I've followed this installation video, but I'm still getting warning messages.
https://youtu.be/_7uBCbVcMlE?si=_iqJB02Psodd_HzE

[mykeehu]

Same here: #5 (comment)
But I have no inference_core_nodes warning.

[mykeehu]

Here's the solution from ChatGPT!

Open \ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI_MagicQuill\LLaVA\llava\model\builder.py and replace lines 32-52 with this:

    kwargs = {"device_map": device_map, **kwargs}

    if device != "cuda":
        kwargs['device_map'] = {"": device}

    if load_8bit:
        kwargs['load_in_8bit'] = True
        kwargs['quantization_config'] = None
    elif load_4bit:
        kwargs['load_in_4bit'] = True
        kwargs['quantization_config'] = BitsAndBytesConfig(
            load_in_4bit=True,
            bnb_4bit_compute_dtype=torch.float16,
            bnb_4bit_use_double_quant=True,
            bnb_4bit_quant_type='nf4'
        )
    else:
        kwargs['torch_dtype'] = torch.float16

    if use_flash_attn:
        kwargs['attn_implementation'] = 'flash_attention_2'

    if 'quantization_config' in kwargs:
        kwargs.pop('load_in_4bit', None)

[KLL535]

This thing took half of my VRAM right after starting comfyui, why is that? How do I disable LMM autoload? My memory is not infinite, it's very little.