Problem description

Summary
The official rlinf Docker images are currently incompatible with the new NVIDIA RTX 5090 GPUs due to PyTorch and CUDA version requirements. This issue is a request for guidance and to inquire about official support for this new hardware.

Problem Description
The NVIDIA RTX 5090's architecture requires a newer version of PyTorch (likely 2.8.0 or higher) for compatibility. The current official images are built on PyTorch 2.5.1:

rlinf/rlinf:math-rlinf0.1-torch2.5.1-sglang0.4.4-vllm0.7.1-megatron0.11.0-te2.1
rlinf/rlinf:agentic-openvla-rlinf0.1-torch2.5.1
rlinf/rlinf:agentic-openvlaoft-rlinf0.1-torch2.5.1
Attempted Solution
We have tried to rebuild the Docker images by upgrading PyTorch. However, this has led to significant dependency conflicts with other core libraries in the stack, making it very difficult to create a stable environment.

Questions
For the community: Has anyone successfully built and run the rlinf environment on an RTX 5090? If so, could you share your Dockerfile or the steps you took to resolve the dependencies?
For the maintainers: Are there plans to release official images compatible with the RTX 5090? Any guidance on which dependency versions are known to work with a newer PyTorch would also be incredibly helpful.

Thank you for your fantastic work on this project and for any assistance you can provide.

Before submitting a new issue...

• Have you checked relevant issues, FAQs (https://rlinf-docs.readthedocs.io/en/latest/rst_source/faq.html), or asked the chatbot at the top right corner of the documentation page.