Skip to content

Support CUDA 12 container builds#15747

Open
pzelasko wants to merge 37 commits into
mainfrom
codex/container-cu12
Open

Support CUDA 12 container builds#15747
pzelasko wants to merge 37 commits into
mainfrom
codex/container-cu12

Conversation

@pzelasko
Copy link
Copy Markdown
Collaborator

@pzelasko pzelasko commented Jun 3, 2026

Important

The Update branch button must only be pressed in very rare occassions.
An outdated branch is never blocking the merge of a PR.
Please reach out to the automation team before pressing that button.

What does this PR do ?

  • support CUDA 12
# This CI Dockerfile supports CUDA 13 and CUDA 12 from a single BASE_IMAGE
# build arg. The default is the recommended CUDA 13 image:
#   nvcr.io/nvidia/cuda-dl-base:26.04-cuda13.2-devel-ubuntu24.04
# The current recommended CUDA 12 image is:
#   nvcr.io/nvidia/cuda-dl-base:25.06-cuda12.9-devel-ubuntu24.04
#
# The build derives CUDA_FLAVOR internally from BASE_IMAGE by matching
# "cuda13" or "cuda12" in the image tag. That flavor selects the matching uv
# extra (cu13 or cu12) and CUDA Python package include path. If BASE_IMAGE does
# not contain either token, the build fails early.
#
# Example CUDA 13 H100+ build:
#   docker buildx build -f docker/Dockerfile.ci \
#     --build-arg GPU_TARGET=h100plus .
# Example CUDA 12 A100 build:
#   docker buildx build -f docker/Dockerfile.ci \
#     --build-arg BASE_IMAGE=nvcr.io/nvidia/cuda-dl-base:25.06-cuda12.9-devel-ubuntu24.04 \
#     --build-arg GPU_TARGET=a100 .
  • remove Dockerfile.stable
  • fix aistore dependency so it installs correctly, add smart_open too
  • add optional ffmpeg installation (disabled by default) and torchcodec, in container only (not NeMo depending)

Collection: [Note which collection this PR will affect]

Changelog

  • Add specific line by line info of high level changes in this PR.

Usage

  • You can potentially add a usage example below
# Add a code snippet demonstrating how to use this 

GitHub Actions CI

The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.

The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR.
To re-run CI remove and add the label again.
To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".

Before your PR is "Ready for review"

Pre checks:

  • Make sure you read and followed Contributor guidelines
  • Did you write any new necessary tests?
  • Did you add or update any necessary documentation?
  • Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
    • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

  • New Feature
  • Bugfix
  • Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.

Additional Information

  • Related to # (issue)

chtruong814 and others added 30 commits May 13, 2026 13:40
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
This reverts commit 8c5a48e.

Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Charlie Truong <chtruong@nvidia.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
Use decoder vocab size when generating synthetic TDT transcript labels so duration outputs from the joint are not sampled as labels.

Move CUDA graph compile exception types into cuda_python_utils per review feedback.

Signed-off-by: Piotr Żelasko <pzelasko@nvidia.com>
…led-deps

# Conflicts:
#	docker/Dockerfile.ci
@copy-pr-bot
Copy link
Copy Markdown

copy-pr-bot Bot commented Jun 3, 2026

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

# Conflicts:
#	docker/Dockerfile.ci
#	pyproject.toml
#	uv.lock
@github-actions github-actions Bot added the ASR label Jun 3, 2026
@pzelasko pzelasko changed the title Codex/container cu12 Support CUDA 12 container builds Jun 3, 2026
@pzelasko pzelasko marked this pull request as ready for review June 3, 2026 21:42
@pzelasko pzelasko requested a review from a team as a code owner June 3, 2026 21:42
@pzelasko
Copy link
Copy Markdown
Collaborator Author

pzelasko commented Jun 3, 2026

/ok to test 15815f8

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants