[bug]: Fallback to CPU on ver 3.4.0 on Manjaro Linux with 6900XT

[mcgravier]

Is there an existing issue for this?

• [X] I have searched the existing issues

OS

Linux

GPU

amd

VRAM

16GB

What version did you experience this issue on?

3.4.0

What happened?

After upgrading to 3.4.0post2 InvokeAI falls back to CPU

• I have upgraded package python-pytorch-opt-rocm to 2.1, however it made no difference
• last working version is 3.3.0post3
• downgrading requires clean install of 3.3.0post3, otherwise issue persists
• I'm running Manjaro kernel 6.6.2 with ROCm 5.7.1

Screenshots

No response

Additional context

Downgrading via script to previous version does't seem to work. I had to make a clean install, and move the models, database and pictures manually to the 3.3.0post3

Contact Details

No response

[Millu]

Could you share the server startup logs?

[mcgravier]

Generate images with a browser-based interface [2023-12-01 19:49:18,258]::[InvokeAI]::INFO --> Loaded 0 modules from /home/adam/invokeai/nodes /home/adam/invokeai/.venv/lib/python3.11/site-packages/torchvision/transforms/functional_tensor.py:5: UserWarning: The torchvision.transforms.functional_tensor module is deprecated in 0.15 and will be removed in 0.17. Please don't rely on it. You probably just need to use APIs in torchvision.transforms.functional or in torchvision.transforms.v2.functional. warnings.warn( patchmatch.patch_match: INFO - Compiling and loading c extensions from "/home/adam/invokeai/.venv/lib/python3.11/site-packages/patchmatch". patchmatch.patch_match: ERROR - patchmatch failed to load or compile (libvtkFiltersTexture.so.1: cannot open shared object file: No such file or directory). patchmatch.patch_match: INFO - Refer to https://invoke-ai.github.io/InvokeAI/installation/060_INSTALL_PATCHMATCH/ for installation instructions. [2023-12-01 19:49:24,749]::[InvokeAI]::INFO --> Patchmatch not loaded (nonfatal) [2023-12-01 19:49:26,589]::[uvicorn.error]::INFO --> Started server process [14684] [2023-12-01 19:49:26,589]::[uvicorn.error]::INFO --> Waiting for application startup. [2023-12-01 19:49:26,590]::[InvokeAI]::INFO --> InvokeAI version 3.4.0 [2023-12-01 19:49:26,590]::[InvokeAI]::INFO --> Root directory = /home/adam/invokeai [2023-12-01 19:49:26,590]::[InvokeAI]::INFO --> Using database at /home/adam/invokeai/databases/invokeai.db [2023-12-01 19:49:26,598]::[InvokeAI]::INFO --> GPU device = cpu [2023-12-01 19:49:26,602]::[InvokeAI]::INFO --> Scanning /home/adam/invokeai/models for new models [2023-12-01 19:49:26,774]::[InvokeAI]::INFO --> Scanned 9 files and directories, imported 0 models [2023-12-01 19:49:26,777]::[InvokeAI]::INFO --> Model manager service initialized [2023-12-01 19:49:26,788]::[InvokeAI]::INFO --> Pruned 0 finished queue items [2023-12-01 19:49:26,791]::[InvokeAI]::INFO --> Cleaned database [2023-12-01 19:49:26,791]::[uvicorn.error]::INFO --> Application startup complete. [2023-12-01 19:49:26,791]::[uvicorn.error]::INFO --> Uvicorn running on http://127.0.0.1:9090 (Press CTRL+C to quit)

[Operator21]

Same problem on Debian 12 running with RX6750XT, after updating to 3.4.0post2 invokeai is running CPU only, Rocm worked fine before the update.

[Millu]

Try running this from the developer console:

pip install "torch==2.1.0+rocm5.6" "torchvision==0.16.0+rocm5.6" "requests~=2.28.2" --force-reinstall --extra-index-url https://download.pytorch.org/whl/rocm5.6

[Core420]

That worked for me, thx

[mcgravier]

Try running this from the developer console:

pip install "torch==2.1.0+rocm5.6" "torchvision==0.16.0+rocm5.6" "requests~=2.28.2" --force-reinstall --extra-index-url https://download.pytorch.org/whl/rocm5.6

Thank you, this worked!