Once used, Model Speedup cannot be deactivated

thefoxfarmer commented 3 months ago

Describe the bug

Once a sampling workflow has used the Model Speedup node once, removing the node from the workflow does not have the effect of removing the change to the sampling process.

Your environment

macOS 14.5

OneDiff git commit id

4b35dc2a8d53c3122862b51cc69c66f59738b407

OneFlow version info if you have installed oneflow

(venv) (floppy onediff/ 0 5177)$ python -m oneflow --doctor
libibverbs not available, ibv_fork_init skipped
path: ['/home/ff/ComfyUI/venv/lib/python3.10/site-packages/oneflow']
version: 0.9.1.dev20240720+cu118
git_commit: dcaba7d
cmake_build_type: Release
rdma: True
mlir: True
enterprise: False

How To Reproduce

Insert the Model Speedup node into a working workflow.
Execute the workflow.
Remove the Model Speedup node from the workflow.
Execute the workflow again.

The complete error message

There is no error message. The effect of the Model Speedup node continues, though the node is gone.

Additional context

This would seem like something that one couldn't complain about if this were perfect. (Why remove it?) I don't know how close to perfect it is, but I know it is not perfect, so I need to be able to turn it off and I would prefer not to have to restart ComfyUI to do that. (Even if it did seem to be perfect, I'd want to turn it off to A/B test it now and then.)

strint commented 3 months ago

@ccssu Maybe know about this

ccssu commented 3 months ago

Please set the inplace parameter to False ，then remove this node from the workflow, and its impact on the sampling process will be removed. @thefoxfarmer

siliconflow / onediff