re-evaluating prompt makes comfyui crash

[elemich]

Custom Node Testing

• I have tried disabling custom nodes and the issue persists (see how to disable custom nodes if you need help)

Your question

i'm on colab free(12gb ram,16gb vram),fresh install at every run (models and persistent stuff are on GoogleDrive).
i'm using :

wan2.1 rapid gguf model,about 7,5 gb
umt5-xxl-encoder-Q3_K_M.gguf (2,85gb)
a lora (600mb)
the wan vae (circa 2-300 mb)

making a imageToVideo.i can rerun the workflow cyclically a lot of time with a plain prompt,
but if the prompt needs to be reevaluated (for example if it contains some {|} for random choice),comfyui crashes.
in the terminal i have a very compact ram/vram resource viewer: ram is 46% and vram is 74%.

now if i run main.py with no vram parameters,comfyui crashes.with --highvram it launches a popup but it not crashes,and in this case, if i "unload models and execution cache" from the comfymenu and restart the workflow i can wait a little and the workflow restarts without error until the next reevaluation of the prompt.

is this an error? how can i solve that?maybe --highvram with "UnloadModels and execution cache"node will fix that issue? how can i "unload models and execution cache" from a node after one generation automatically?

can someone explain me the precise problem?

Logs

Other

No response

[rattus128]

You can get the effect of unloading cache at every time with the --cache-none startup argument.

[elemich]

But it used to work fine, but it's only been this way for a few days now. Can you tell me why? I think it's a bug.

[rattus128]

But it used to work fine, but it's only been this way for a few days now. Can you tell me why? I think it's a bug.

If you are doing model reuse flows or upscaling we just fixed a VRAM OOM. Please update both comfy core and ComfyUI-GGUF to the latest git versions.

Otherwise if you still have issues we need a log and workflow to take a bug report.

More is more with the bug reports. If in doubt paste it all.