Skip to content

[WIP] don't review#2420

Draft
shuyixiong wants to merge 15 commits into
NVIDIA-NeMo:mainfrom
shuyixiong:shuyix/trtllm
Draft

[WIP] don't review#2420
shuyixiong wants to merge 15 commits into
NVIDIA-NeMo:mainfrom
shuyixiong:shuyix/trtllm

Conversation

@shuyixiong
Copy link
Copy Markdown
Contributor

What does this PR do ?

Add a one line overview of what this PR aims to accomplish.

Issues

List issues that this PR closes (syntax):

Usage

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

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 run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
  • Did you add or update any necessary documentation? Visit our Document Development Guide for how to write, build and test the docs.

Additional Information

  • ...

@copy-pr-bot
Copy link
Copy Markdown

copy-pr-bot Bot commented May 6, 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.

@github-actions github-actions Bot added the Documentation Improvements or additions to documentation label May 6, 2026
ricklamers and others added 11 commits May 12, 2026 00:56
Add the TRT-LLM PyTorch backend integration needed to validate GRPO with draft-target speculative decoding, Mongo rollout logging, and weight-sync timing on AIHub.
Remove the experimental speculative-decoding hooks (config, worker
plumbing, spec_token_origins propagation, specdec exemplar config)
and the MongoDB logger backend that were bundled with the TRT-LLM
backend introduction. The TRT-LLM PyTorch backend itself is unchanged.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Signed-off-by: shuyixiong <219646547+shuyixiong@users.noreply.github.com>
TMP commit (will amend or squash later). Layers TRT-LLM-specific
wiring on top of the CUDA 13 squash from NVIDIA-NeMo#2332:

  - pyproject.toml: pin trtllm extra to tensorrt_llm==1.3.0rc13
  - virtual_cluster.py: add PY_EXECUTABLES.TRTLLM
  - ray_actor_environment_registry.py: add TRTLLM_EXECUTABLE
    (with NEMO_RL_PY_EXECUTABLES_SYSTEM=1 fallback)
  - docker/Dockerfile: add SKIP_TRTLLM_BUILD arg + HPCX env exposure
    (cuda-compat removed — cu13.2 base image already provides it)
  - docs/docker.md: SKIP_TRTLLM_BUILD usage docs

Known TODOs before this can be promoted:
  - `uv lock` likely fails: tensorrt-llm 1.3.0rc13 has no cp313 wheel.
    Either move trtllm to its own workspace (cp312) or relax pin.
  - Build + smoke test on a GB200 node not yet validated.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Signed-off-by: shuyixiong <219646547+shuyixiong@users.noreply.github.com>
Signed-off-by: shuyixiong <219646547+shuyixiong@users.noreply.github.com>
Signed-off-by: shuyixiong <219646547+shuyixiong@users.noreply.github.com>
Signed-off-by: shuyixiong <219646547+shuyixiong@users.noreply.github.com>
Signed-off-by: shuyixiong <219646547+shuyixiong@users.noreply.github.com>
Signed-off-by: shuyixiong <219646547+shuyixiong@users.noreply.github.com>
Signed-off-by: shuyixiong <219646547+shuyixiong@users.noreply.github.com>
Signed-off-by: shuyixiong <219646547+shuyixiong@users.noreply.github.com>
Signed-off-by: shuyixiong <219646547+shuyixiong@users.noreply.github.com>
Signed-off-by: shuyixiong <219646547+shuyixiong@users.noreply.github.com>
Signed-off-by: shuyixiong <219646547+shuyixiong@users.noreply.github.com>
Signed-off-by: shuyixiong <219646547+shuyixiong@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Documentation Improvements or additions to documentation

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants