feat: Add SGLang rollout backend and tests

[RolaoDenthu]

What does this PR do ?

Add comprehensive test coverage for SGLang generation backend, including functional tests, unit tests, and nightly tests.

• Functional Test (tests/functional/grpo_sglang.sh): Quick validation of SGLang-based GRPO training
• Unit Tests (tests/unit/models/generation/test_sglang_generation.py): unit tests covering:
  • Basic configuration validation
  • Policy generation and tensor parallelism
  • Worker seed behavior for RLHF diversity
  • HTTP server direct API access
  • Weight updates with DTensor policy (colocated mode)
  • Prefix cache reset after weight updates
• Nightly Test (tests/test_suites/llm/grpo-qwen3-0.6b-1n8g-sglang.sh): End-to-end convergence test for SGLang backend

Usage

• You can potentially add a usage example below

# Run functional test
uv add coverage
bash tests/functional/grpo_sglang.sh

# Run unit tests
uv sync --extra sglang --group test
uv run python -m pytest tests/unit/models/generation/test_sglang_generation.py -v --sglang-only

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.

Summary by CodeRabbit

Release Notes

• New Features

  • Distributed generation engine using SGLang backend with HTTP weight streaming and multi-GPU support.

• Configuration

  • New YAML configuration templates for SGLang-based experiments with customizable generation parameters.

• Tests

  • Comprehensive test coverage for SGLang generation, including tensor parallelism, batching, and dynamic weight updates.

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[github-actions]

⚠️ File Consistency Check

Check based on commit: d037f71 (PR #1674 from add-tests)

⚠️ DTensor Policy Worker Synchronization Warning

The file nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py was modified in this PR, but nemo_rl/models/policy/workers/dtensor_policy_worker.py was not updated.

Why this matters:
These files contain related DTensor policy worker implementations that should be kept synchronized to ensure consistency across different versions.

Action required:

• Please review if the changes in nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py should also be applied to nemo_rl/models/policy/workers/dtensor_policy_worker.py
• Update nemo_rl/models/policy/workers/dtensor_policy_worker.py if necessary to maintain consistency
• If the files are intentionally different, please add a comment in the PR explaining why

Files to check:

• Modified: nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py
• Not modified: nemo_rl/models/policy/workers/dtensor_policy_worker.py

---

_{This check ensures that related file implementations remain synchronized across the codebase. If you believe this warning is incorrect or the files should intentionally differ, please add a comment explaining the reasoning.}

[RolaoDenthu]

⚠️ File Consistency Check

Check based on commit: d037f71 (PR #1674 from add-tests)

⚠️ DTensor Policy Worker Synchronization Warning

The file nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py was modified in this PR, but nemo_rl/models/policy/workers/dtensor_policy_worker.py was not updated.

Why this matters: These files contain related DTensor policy worker implementations that should be kept synchronized to ensure consistency across different versions.

Action required:

• Please review if the changes in nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py should also be applied to nemo_rl/models/policy/workers/dtensor_policy_worker.py
• Update nemo_rl/models/policy/workers/dtensor_policy_worker.py if necessary to maintain consistency
• If the files are intentionally different, please add a comment in the PR explaining why

Files to check:

• Modified: nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py
• Not modified: nemo_rl/models/policy/workers/dtensor_policy_worker.py

This check ensures that related file implementations remain synchronized across the codebase. If you believe this warning is incorrect or the files should intentionally differ, please add a comment explaining the reasoning.

SGLang supports the weight update function only for DTensor v2, while the original DTensor worker does not. Therefore, this change is intentionally applied only to dtensor_policy_worker_v2.py.

[guyueh1]

⚠️ File Consistency Check

Check based on commit: d037f71 (PR #1674 from add-tests)

⚠️ DTensor Policy Worker Synchronization Warning

The file nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py was modified in this PR, but nemo_rl/models/policy/workers/dtensor_policy_worker.py was not updated.
Why this matters: These files contain related DTensor policy worker implementations that should be kept synchronized to ensure consistency across different versions.
Action required:

• Please review if the changes in nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py should also be applied to nemo_rl/models/policy/workers/dtensor_policy_worker.py
• Update nemo_rl/models/policy/workers/dtensor_policy_worker.py if necessary to maintain consistency
• If the files are intentionally different, please add a comment in the PR explaining why

Files to check:

• Modified: nemo_rl/models/policy/workers/dtensor_policy_worker_v2.py
• Not modified: nemo_rl/models/policy/workers/dtensor_policy_worker.py

This check ensures that related file implementations remain synchronized across the codebase. If you believe this warning is incorrect or the files should intentionally differ, please add a comment explaining the reasoning.

SGLang supports the weight update function only for DTensor v2, while the original DTensor worker does not. Therefore, this change is intentionally applied only to dtensor_policy_worker_v2.py.

I think this is fine to ignore, the API is defined in the base worker as "not implemented", so there is no risk that calling this method with dtensor (v1) object will cause a crash, but it will be caught by a not implemented error.