When will we be getting Quanto support for Wan 2.1?

[ukaprch]

The diffusers library for quantizers currently doesn't contain an entry for Quantro:

https://github.com/huggingface/diffusers/tree/main/src/diffusers/quantizers

Isn't this needed to perform requantization on a quantized Transformer for WAN 2.1?

Currently we can't do this due to missing Quanto quantizer after we've quantized and stored a Transformer:

print('Quantize transformer') class QuantizedWanTransformer3DModel(QuantizedDiffusersModel): base_class = WanTransformer3DModel transformer = QuantizedWanTransformer3DModel.from_pretrained( "./wan quantro T2V 14B Diffusers/basemodel/wantransformer3dmodel_qint8" ).to(dtype=dtype)

[DN6]

@ukaprch Can you please share a code example for how you're trying to load the model so that I can understand the issue? Based on what you've posted here it looks like you're trying to load via a wrapped model which shouldn't be necessary. Additionally can you share a link to the checkpoint you're trying to load?

[ukaprch]

We can close this issue.