Load Checkpoint (Quantized, Simple)

The "Simple" variant of Load Checkpoint (Quantized) is the pack's answer to people who just want the thing to load. It strips out the quant_format and kernel_backend dropdowns entirely and runs the full loader with format set to auto and backend set to pytorch - the source is literally a one-line call to the non-Simple class. You pick the checkpoint, it detects each layer's quant format on its own, and out comes the same triple output you'd get from stock Load Checkpoint.

This is the node to use when you downloaded a quantized combined checkpoint from silveroxides' HuggingFace - diffusion model, text encoder, and VAE all in one file - and you don't care about the mechanics. If your file is split into a separate UNET and text encoder instead, the separate loaders are the right tools; this one is strictly for combined checkpoints.

The inputs that matter

• ckpt_name - the combined checkpoint from your checkpoints folder.
• disable_dynamic and low_memory - the pack's shared loading toggles (low_memory tooltip: fast low-impact loading, or ComfyUI's default when off).

Outputs are MODEL, CLIP, and VAE - same wiring as the stock loader, so nothing downstream changes.

Installing it

ComfyUI Manager → search "ComfyUI-QuantOps", or:

cd ComfyUI/custom_nodes
git clone https://github.com/silveroxides/ComfyUI-QuantOps

Restart and let requirements.txt pull in unifiedefficientloader>=0.5.2.

The catch

"Auto" covers the pack's recognized layouts - int8, tensorwise int8, the fp8 variants, mxfp8, nvfp4 - and nothing else. A checkpoint quantized some other way won't load just because this node is called Simple. Also, the pack is deprecated: int8 ConvRot is native in ComfyUI (v0.27.0+) and the author has stopped maintaining it, so on a current ComfyUI the built-in loader probably already handles your file. Keep this node around for the older combined checkpoints made with the convert_to_quant toolchain; don't build a fresh stack on it.