[rollout][DONT-MERGE] Test TRTLLM PPO CI workflow

[hchings]

What does this PR do?

Add concise overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review.

Checklist Before Starting

• Search for similar PRs. Paste at least one query link here: ...
• Format the PR title as [{modules}] {type}: {description} (This will be checked by the CI)
  • {modules} include fsdp, megatron, sglang, vllm, rollout, trainer, ci, training_utils, recipe, hardware, deployment, ray, worker, single_controller, misc, perf, model, algo, env, tool, ckpt, doc, data, cfg, reward
  • If this PR involves multiple modules, separate them with , like [megatron, fsdp, doc]
  • {type} is in feat, fix, refactor, chore, test
  • If this PR breaks any API (CLI arguments, config, function signature, etc.), add [BREAKING] to the beginning of the title.
  • Example: [BREAKING][fsdp, megatron] feat: dynamic batching

Test

For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc.

API and Usage Example

Demonstrate how the API changes if any, and provide usage example(s) if possible.

# Add code snippet or script demonstrating how to use this

Design & Code Changes

Demonstrate the high-level design if this PR is complex, and list the specific changes.

Checklist Before Submitting

Important

Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review.

• Read the Contribute Guide.
• Apply pre-commit checks: pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always
• Add / Update the documentation.
• Add unit or end-to-end test(s) to the CI workflow to cover all the code. If not feasible, explain why: ...
• Once your PR is ready for CI, send a message in the ci-request channel in the verl Slack workspace. (If not accessible, please try the Feishu group (飞书群).)
• If your PR is related to the recipe submodule, please also update the reference to the submodule commit via git submodule update --remote or cd recipe && git pull origin main.

[gemini-code-assist]

Code Review

This pull request introduces support for TensorRT-LLM as a rollout engine. The changes are extensive, touching Dockerfiles, documentation, configuration, and core Python code for workers and trainers. The implementation adds a new trtllm rollout worker, server, and replica, integrating with the existing Ray-based infrastructure. The use of IPC for weight updates and careful management of placement groups for co-location of training and inference workers are notable. Overall, the changes seem well-designed to integrate TRTLLM, but there are a couple of points in the Dockerfile that could be improved for better reliability and maintainability.

[gemini-code-assist]

The wget command on line 17 downloads nvshmem from a URL that contains /secure/. This often indicates that the resource is behind a login or requires some form of authentication, which can cause the Docker build to fail in automated CI environments. It's better to either host this dependency in a more accessible location or to use a base image that already includes it if possible. If this is a public but temporary URL, it's still a risk for long-term build reproducibility.

[gemini-code-assist]

Installing verl from git and then immediately uninstalling it is a confusing pattern for managing dependencies. While this might be intended to cache the dependencies in a Docker layer, it makes the Dockerfile harder to understand and maintain. A more explicit approach would be better. For example, you could clone the repository, generate a requirements.txt from setup.py, and then install dependencies from that file. This would make the intent clear and separate dependency installation from the package installation.