[BREAKING][worker, rollout, vllm] feat: implement vLLM colocated training-inference rollout with process separation

[jianjunzhong]

What does this PR do?

Refactor vLLM co-located training-inference rollout from single-process to multi-process architecture. This refactoring separates training and inference into different processes, enabling better resource isolation and paving the way for future checkpoint-engine integration (in roadmap #3624).

Key Changes:

• Transform vLLMAsyncRollout into ServerAdapter - a client-side adapter that communicates with the inference executor
• Remove ExternalZeroMQDistributedExecutor and use MultiprocExecutor as the inference backend
• Implement CUDA IPC-based weight updates via ZeroMQ for efficient parameter synchronization between training and inference processes

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

This refactoring maintains full backward compatibility with existing vLLM rollout APIs. No changes are required to user code.

Key API Components:

• ServerAdapter (replaces vLLMAsyncRollout):
  • Acts as client-side adapter for communicating with inference executor
  • Manages CUDA IPC-based weight updates
  • Provides same interface as previous vLLMAsyncRollout class

Design

Architecture Overview

1. Before (Single-Process Architecture)

• Single-Process Design

In the original AsyncActorRolloutRefWorker, the training engine and inference engine shared the same process. The vLLM inference engine directly received weight updates through parameter passing.

• Communication Architecture

ExternalZeroMQDistributedExecutor acts as a client, sending instructions to all AsyncActorRolloutRefWorker inference engines via ZMQ to execute operations like init_worker, load_model, init_device, and generate. Operations like wake_up, sleep, and weight updates were executed directly in vLLMAsyncRollout without going through ExternalZeroMQDistributedExecutor.

2. After (Multi-Process Architecture):

• Multi-Process Design

Transform vLLMAsyncRollout into ServerAdapter, serving as a client for communicating with the inference engine (AsyncLLM). Weight updates are based on CUDA IPC, passing through ZeroMQ to the inference engine.

• Communication Architecture

Deprecate the original ExternalZeroMQDistributedExecutor class and directly use vllm's MultiprocExecutor by passing distributed_executor_backend = "mp". All inference engine operations are uniformly broadcast to all inference workers through MultiprocExecutor's RPC Broadcast MQ.

Convergence test

1. GPU

• model: Qwen3-VL-30B-A3B-Instruct
• dataset: geo3k
• GPU: 4*8 H100

2. NPU

• model: Qwen3-8B
• dataset: gsm8k
• NPU: 1*8 910C
• Ascend HDK: 25.3.RC1.2
• CANN: 8.3.RC2
• vLLM-Ascend: 0.13.0rc1
• training backend: FSDP
  

Performance test: update weights

• CUDA IPC bucket_size: 2GB
• GPU: H100, ConnectX-7 400 Gbps (InfiniBand)

Model	#GPU	Parallelism	Time
Qwen3-VL-30B-A3B-Instruct	TP2,EP8	4*8	5s
DeepSeek-V3.1-Terminus	TP8, PP16, EP8	16*8	120s
DeepSeek-V3.1-Terminus	TP16,PP16	32*8	80s

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).)

[pengwu22]

hi. thank you for the contribution! just one comment: could you abstract the zmq+ipc communication channel out, and add some corresponding unittests please?