lightx2v

/

Qwen-Image-Lightning

like 798

When I try to use torch.compile with that LoRA, it fails complaining about recompilation limits reached.
Could it be caused by a lack of VRAM (and this being FP32) ?

Or could be an issue of ComfyUI with Qwen-Image and LoRA for it in general...
I opened issues:
ComfyUI: https://github.com/comfyanonymous/ComfyUI/issues/9289
KJ Nodes: https://github.com/kijai/ComfyUI-KJNodes/issues/363

EDIT: I tried to resave this lora as BF16 (except the INT64 parts ofc), and same issue