Motivation

Torch 2.11 will set Cuda 13.0 as the default version. So it's necessary for SGLang to shift default cuda version from Cuda 12.9 to Cuda 13.0

We plan to do the migration after release of 0.5.10 version for SGLang

Checklist

It involves multiple steps to fully shift all the Cuda dependencies from 12.9 to 13.0:

• Ramp up 13.0 environment on all CI runners, and modify CI-related scripts/workflows [CI] Add per-job uv venv isolation and upgrade CI version to Cuda 13 #23119
  • Change default cuda-python to cuda 13.0 in pyproject.toml
  • Change packages to cuda 13 version in ci_install_dependency.sh
• After CI gets stable on Cuda 13, shift docker images to cuda 13 [Docker] Prep for torch 2.11: cu129 fix, image validator, dep cleanup #23593
  • Modifty Dockerfiles (sglang, sgl-kernel)
  • Modify docker release workflows, so that the default image tag is pointed to cu13 images. (release-docker.yml, release-docker-dev.yml)
• Redirect default sglang wheel to Cuda 13 version [CI] Publish nightly sglang wheel under both cu129 and cu130 indexes #24176 [Misc] Redirect default sglang nightly wheel to cuda 130 #24183
  • Change default cuda version in update_nightly_whl_index.py
  • Build up a support matrix (cu129/cu130) for nightly sglang wheel in release-pypi-nightly.tml
• Redirect default sglang-kernel wheel to Cuda 13 version [sgl-kernel] Prep for torch 2.11 upgrade and switch PyPI default to cu130 #24162
  • Modify release-whl-kernel.yml, so that the cu130 wheels will be uploaded to pypi instead of cu129 wheels
  • Modify update_kernel_whl_index.py, so that the default cuda version on whl index is shifted to cuda 13.0
  • Bump a new version for sglang kernel, to test its functionality on cuda-13 chore: bump sgl-kernel version to 0.4.2 #24170
  • Update document Update kernel installation instructions after shifting default cuda to 13 #24181
• Update documents https://docs.sglang.io/get_started/install.html [Doc] Fix instruction on Cuda 13 environments #24516
• Upgrade Torch to 2.11 [Dependency] Upgrade to Torch 2.11.0 #21247
• Upgrade to Cuda 13.1 （optional, might depend on the next torch version)

All these steps should be finished in the listed order.
Cuda 12.9 image/kernels will still be under maintenance. It's just they are not the default option.

Related resources

• Torch 2.11 release blog https://pytorch.org/blog/pytorch-2-11-release-blog/
• [RFC] Release cuda 13.0 docker image as default #20486

No response