Backend impacted

The PyTorch implementation

Operating system

Linux

Hardware

GPU with CUDA

Description

Summary

Running the provided Docker setup fails on NVIDIA GeForce RTX 5090 (Blackwell, sm_120).
The container exits with a PyTorch warning about unsupported CUDA capability and then crashes with:
RuntimeError: CUDA error: no kernel image is available for execution on the device.

Environment

• Host OS: Windows (Docker Desktop + WSL2)
• Container base image: nvcr.io/nvidia/cuda:12.4.1-runtime-ubuntu22.04
• GPU: NVIDIA GeForce RTX 5090 (Blackwell, compute capability sm_120)
• Docker Compose: uses driver: nvidia and capabilities: [gpu]

Steps to reproduce

1. Clone repo
2. docker compose up -d --build
3. Observe container logs and crash

Expected behavior

Container starts successfully and uses the GPU for inference.

Actual behavior

Container exits during model initialization. Log shows:

NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90.
...
RuntimeError: CUDA error: no kernel image is available for execution on the device

### Extra information

N/A

### Environment

Docker compose