NVIDIA 5060 incompatibility

[Spartbrine]

I'm trying to make a workflow at n8n ussing OLLAMA & XTTS which uses deepspeed, but it make me an incompatibility with my drivers and more problems with deepspeed & other libraries, should I use this in my PC with python, Im using all the enviroment at Docker.

[vincere-mori]

RTX 5060 is Blackwell (compute capability 12.0), and most pre-built wheels for PyTorch and DeepSpeed don't include sm_120 yet. So the GPU exists but the CUDA extensions can't target it.

First thing to check: does XTTS actually require DeepSpeed or does it just use it if available? You can usually disable it in the TTS config or by setting DS_BUILD_OPS=0 before starting, which prevents DeepSpeed from loading its CUDA ops at all. XTTS will run slower but it'll run.

If you specifically need full DeepSpeed acceleration, you'd have to build PyTorch from source with TORCH_CUDA_ARCH_LIST="12.0" and then build DeepSpeed against that. In Docker that means rebuilding the base image, which is a lot of work.

Start with DS_BUILD_OPS=0 and see if that unblocks you.