Regarding the loading of the model

by someshijun - opened Feb 21

Feb 21

Hello, I'd like to take the liberty of asking. If I use the latest INT8 (W8A8) model loading node for loading, how should I set the model_type? Can you provide a screenshot for reference? Thank you.

Bedovyy

Owner Feb 21

The latest ComfyUI-Flux2-INT8 has on-the-fly option, and if you enable it, the model you selected will be quantized on-the-fly as model_type.

In other word, you don't need to set model_type (keep it as default - Flux2 afaik) when you use prequantized model - just disable on-the-fly option.

someshijun

Feb 22

This comment has been hidden (marked as Off-Topic)

someshijun changed discussion status to closed Feb 22

someshijun changed discussion status to open Feb 22

Winnougan

Feb 25

It doesn't work! Can you provide examples?

Bedovyy

Owner Feb 25

You can use int8 model (Tensorwide) model with custom_node https://github.com/BobJohnson24/ComfyUI-Flux2-INT8
If you want to use int8rowwise model, You should check the custom node is in Quip branch.

~/Servers/ComfyUI/custom_nodes/ComfyUI-Flux2-INT8$ git branch
* Quip
  main

Then, use the model as below.

Winnougan

Mar 12

•

edited Mar 12

Thanks

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment