Open ethanfel opened 7 months ago
Cannot verify. https://github.com/comfyanonymous/ComfyUI/commit/a28a9dc83684624ee2167c0b92d976bb68f2c606 seems to work fine.
Cannot verify. comfyanonymous/ComfyUI@a28a9dc seems to work fine.
Added a simple WF to illustrate the issue.
start comfyui -> run -> error.
but
bypass layer diff -> run (work) -> enable layer diff -> run (work until the model is changed or comfyui restarted)
the bug is still present in the later comfyui commit.
What happened?
RuntimeError: "addmm_implcpu" not implemented for 'Half' in the recent comfyui commits.
Steps to reproduce the problem
running wf
What should have happened?
The mm call (matrix multiplication) requires float32, but it gets a half type (float16)
Commit where the problem happens
ComfyUI: after d7897ff ComfyUI-layerdiffuse:
Sysinfo
Linux:RTX4090
Console logs
Workflow json file
none
Additional information
No response