What does this PR do?

When using verl+megatron+sglang, we observed extremely unusual phenomena. We found that some abnormal processes were always running on GPU0, occupying VRAM.

When Ray launches each process, it sets CUDA_VISIBLE_DEVICES for each process. For example, rank 1 is set to GPU 1, and rank 2 is set to GPU 2. However, the problem is that from the perspective of process rank 2, it treats GPU 2 as its own GPU 0 (index=0). This leads to rank 2 actually sending a GPU ID of 0 when transferring Tensors across processes. Consequently, every Worker in SGlang receives a Tensor with an incorrect GPU ID, causing abnormal memory usage on GPU 0,at the same time, this also causes unnecessary memory copies, slowing down the update_weights process.

PR addresses the issue of transferring Tensors across processes. This requires a patch to be applied both during serialization in the sending process and deserialization in the receiving process. Verl hasn't implemented this. Referencing PR and the implementation in slime, we resolved this problem.
We also modified the Sglang.

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
  • 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

• 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.