Plans on LyCORIS support?

[JakobLS]

Do you have any plans on supporting LyCORIS weights as well, a part from current LoRA?

When I try to apply them on the INT4 pipeline as follows

transformer = NunchakuFluxTransformer2dModel.from_pretrained("mit-han-lab/svdq-int4-flux.1-dev")
t2i_pipe = AutoPipelineForText2Image.from_pretrained(
    "black-forest-labs/FLUX.1-dev",
    transformer=transformer,
    torch_dtype=torch.bfloat16
)

# Apply LoCORIS weights
lycoris_net, _ = create_lycoris_from_weights(1, weight_path, t2i_pipe.transformer)
lycoris_net.apply_to()

the code completes without erroring out:

2024-12-11 16:41:20|[LyCORIS]-INFO: Loading Modules from state dict...
2024-12-11 16:41:20|[LyCORIS]-INFO: 0 Modules Loaded

But 0 modules are loaded, and I get no effect at all on the resulting images.

When I apply the weights on the bf16 model, 342 modules are loaded and I do get an effect. It looks like the modules are not properly loaded onto the pipeline.

Or perhaps this is more on LYCORIS's side?

[lekima]

Would love to have this supported as well 🙏