cant cancel job , qwen image training w11

[metamountain]

[WORKER] TOOLKIT_ROOT: C:\Ai-Toolkit\AI-Toolkit [WORKER] Cron worker started with interval: 1000 ms [UI] ▲ Next.js 15.1.7 [UI] - Local: http://localhost:8675 [UI] - Network: http://192.168.188.30:8675 [UI] [UI] ✓ Starting... [UI] ✓ Ready in 2.3s [UI] (node:28484) Warning: --localstorage-file was provided without a valid path [UI] (Use node --trace-warnings ... to show where the warning was created) [UI] state { loading: true, gpuData: null, error: null, lastUpdated: null } [WORKER] Stopping job 781aa97a-d7c0-4fde-9de1-4e1bdceee775 on GPU(s) 0 [WORKER] Stopping job 781aa97a-d7c0-4fde-9de1-4e1bdceee775 on GPU(s) 0 [WORKER] Stopping job 781aa97a-d7c0-4fde-9de1-4e1bdceee775 on GPU(s) 0 [WORKER] Stopping job 781aa97a-d7c0-4fde-9de1-4e1bdceee775 on GPU(s) 0 . . . . .

cant cancel jobs. had to edit aitk_db.db with db browser and set job from running to stopped....

have problems with prisma too:

Prisma schema loaded from prisma\schema.prisma Error: EPERM: operation not permitted, rename 'C:\Ai-Toolkit\AI-Toolkit\ui\node_modules.prisma\client\query_engine-windows.dll.node.tmp26644' -> 'C:\Ai-Toolkit\AI-Toolkit\ui\node_modules.prisma\client\query_engine-windows.dll.node'

This should be a problem with permissions....

Delete locked/temp files

Fix permissions (grant full control)

Had to fix it with some powershell script claude.ai wrote for me....

Another problem: Samples before training took 582s/it and cpu was involved (CPU 50% GPU 40%)... Dont know why this happens. So 1 hour for 5 samples seems ill: Qwen 3090

Thanks anyway. I think i have to give up, spent 12 hours with no success... Extremely frustrating was the model downloading which started 3 times for 80 gb... model caches were not correctly retrieved , probably because i stopped training....

Finally i got it running (test file: only 11 images 512*512)

Svne 2: 1%| | 9/1000 [05:33<9:20:51, 33.96s/it, lr: 2.0e-04 loss: 1.359e-01]

Still no gpu load and slow as hell...

Thanks anyway , the idea is super and the interface looks great !

Would be nice if a training loss graph could be implemented. Model structure should be more transparent: Model folder in Ai Toolkit or manual input of path would be nice (kohya style).... Most users have all the models in their comfyui folders.... of course not the Qwen safetensor ones :) But it should be easier to download models separately and put them to documented folder structure. Correct Models download path should be displayed in interface as link. Loading a config file from interface should be possible!