Latest update broke patch model patcher + torch compile

[i-am-flashh]

I tried to regenerate the image with a previous workflow that was using (patch model patcher + torch compile) but the lora is not being applied. The image is being generated without lora.

When i remove the patch model patcher + torch compile nodes, the lora is applied.

It occasionally works , if i restart.

Note: I recently updated comfyui and all custom nodes,

[frankyifei]

Same

[15053285989]

same

[jepjoo]

Same here, running the exact same workflow from 2 weeks ago and getting an image where LoRA isn't applied.

This is a really awesome feature, hoping it can get fixed.

[kijai]

I have tracked this down to this commit, haven't figured out yet why it happens though: comfyanonymous/ComfyUI@79d5cea

[kijai]

Okay so it works again, I'm not sure if this breaks something else now though, but basic workflow with LoRA and compile runs at the compiled speed with LoRA, and I can swap the LoRA just fine.

[akedia]

Looks like the fix make the lora switch works, but the speed after switch is degraded back to no complie.

[frankyifei]

It is worse than no compile, for me it is 10 times, and I have to add --low vram, I wonder if it is loading model in ram instead or vram akedia ***@***.***> 于 2024年12月18日周三 22:43写道：

…

Looks like the fix make the lora switch works, but the speed after switch is degraded back to no complie. — Reply to this email directly, view it on GitHub <#151 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB52SO5TGBSMZHAI7JY2ISD2GFRGBAVCNFSM6AAAAABTAQ34M6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKNJRGE3DKNRSG4> . You are receiving this because you commented.Message ID: ***@***.***>