Skip to content

feat(cli): use CUDA 12.9 as the workstation default - #2638

Merged
JackRao123 merged 2 commits into
mainfrom
jackrao/workstation-b200-cuda129
Sep 1, 2026
Merged

feat(cli): use CUDA 12.9 as the workstation default#2638
JackRao123 merged 2 commits into
mainfrom
jackrao/workstation-b200-cuda129

Conversation

@JackRao123

@JackRao123 JackRao123 commented Sep 1, 2026

Copy link
Copy Markdown
Contributor

Summary

  • change the default training workstation image to nvidia/cuda:12.9.1-devel-ubuntu24.04
  • retain CUDA 13.0.3 for B300/GB300
  • cover default and accelerator-specific image selection with explicit tests

Why

HybridEP runtime-JIT kernels require CUDA 12.9's CCCL/libcu++ headers. The current workstation default is CUDA 12.8.1, so the pinned DeepEP HybridEP kernel fails to compile at startup on B200.

This also covers the devbox-up workflow, which launches boxes through truss train workstation without an image override.

Companion production-image PR: https://github.com/basetenlabs/trainers/pull/1262

Validation

  • uv run ruff format truss/cli/train/workstation.py truss/tests/cli/train/test_workstation.py
  • uv run ruff check truss/cli/train/workstation.py truss/tests/cli/train/test_workstation.py
  • uv run pytest truss/tests/cli/train/test_workstation.py -v (26 passed)

@JackRao123 JackRao123 changed the title feat(cli): use CUDA 12.9 for B200 workstations feat(cli): use CUDA 12.9 as the workstation default Sep 1, 2026
@JackRao123
JackRao123 merged commit 4bb33f9 into main Sep 1, 2026
36 checks passed
@JackRao123
JackRao123 deleted the jackrao/workstation-b200-cuda129 branch September 1, 2026 21:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants