LoRA for Wan2.1-VACE-1.3B

[juanmed]

Hi,

Thanks for open sourcing and sharing your approach. I was trying to get some results but keep getting the error below, and I believe its because I am specifying the use of Wan2.1-VACE-1.3B but the LoRA weights are available only for the Wan2.1-VACE-14B model. Is there a version of Pickford/pickstyle for the 1.3B model?

Traceback (most recent call last):
  File "/home/juan/repos/pickstyle/vace/vace_wan_inference.py", line 490, in <module>
    main(args)
  File "/home/juan/repos/pickstyle/vace/vace_wan_inference.py", line 366, in main
    wan_vace.add_lora_to_vace(pretrained_lora_path=args.pretrained_lora_path, lora_rank=args.lora_rank, lora_alpha=args.lora_alpha,
  File "/home/juan/repos/pickstyle/vace/models/wan/wan_vace.py", line 380, in add_lora_to_vace
    load_multi_lora_weights(self.model, pretrained_lora_path, adapter_name)
  File "/home/juan/repos/pickstyle/vace/models/wan/wan_vace.py", line 175, in load_multi_lora_weights
    mod.lora_A[adapter_name].data.copy_(v)
RuntimeError: The size of tensor a (1536) must match the size of tensor b (5120) at non-singleton dimension 1

The reason I am use the 1.3B model is limited VRAM of 12GB (the 14B model is too large to fit and I ket getting OOM).

Thank you!