Ability for multiple workers and support for sm_120 in the CUDA Binaries

[mycotrip]

I have a medium large library, about 12,000 tracks. The ability have multiple workers would improve speed. It seems the GPU is hardly being used and a sustained work-load would improve scanning speed. Also it seems the CUDA Binaries in the worker don't support sm_120. I see this message sporadically in the logs.

2026-01-03 17:56:45
NVIDIA GeForce RTX 5070 with CUDA capability sm_120 is not compatible with the current PyTorch installation.

2026-01-03 17:56:45
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90.

2026-01-03 17:56:45
If you want to use the NVIDIA GeForce RTX 5070 GPU with PyTorch, please check the instructions at [https://pytorch.org/get-started/locally/⁠](https://pytorch.org/get-started/locally/)

My environment
Docker Desktop on Win 11
Latest NVIDIA GameReady Drivers and CUDA Toolkit

[NeptuneHub]

Hi and thanks for the feedback. Multiple worker is already supported: you just have to raise multiple containers on multiple computer.

have a look to docs/architecture.md page to better understand it (mainly the worker need to see all the other component) and in deployment/ you can look already an example of server and worker deployment splitted.

If you use the -nvidia image GPU will be used. For inference is used ONNX, PyTorch https://github.com/NeptuneHub/AudioMuse-AI/blob/main/docs/ARCHITECTURE.md is used for secondary functionality (like memory cleaning).

[mycotrip]

Hey thanks for the reply! I must have missed that part in the architecture.md. I was able to clear the message I posted by updating torch to 2.9 and cuda to 13 in the Worker instance. As we speak I'm almost done with the first run analysis. It has processed about 173 albums in just about 9 hours.

…

On Sun, Jan 4, 2026 at 1:54 AM NeptuneHub ***@***.***> wrote: Hi and thanks for the feedback. Multiple worker is already supported: you just have to raise multiple containers on multiple computer. have a look to docs/architecture.md page to better understand it (mainly the worker need to see all the other component) and in deployment/ you can look already an example of server and worker deployment splitted. If you use the -nvidia image GPU will be used. For inference is used ONNX, PyTorch https://github.com/NeptuneHub/AudioMuse-AI/blob/main/docs/ARCHITECTURE.md is used for secondary functionality (like memory cleaning). — Reply to this email directly, view it on GitHub <#267 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEKEER5F5UNSNKBUTRZO3P34FC2KHAVCNFSM6AAAAACQTTPKOKVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTKNBQGM2DGNQ> . You are receiving this because you authored the thread.Message ID: ***@***.*** com>