chore(trtllm): bump NIXL to v1.3.1

[tanmayv25]

Summary

Bump the trtllm nixl_ref from v1.0.1 to v1.3.1 (one line in container/context.yaml).

Validation

Built the trtllm runtime image locally with nixl_ref=v1.3.1:

• nixl-sys compiles from the v1.3.1 source; nixl / nixl-cu13 install at 1.3.1; NIXL initializes cleanly (NixlTransferAgent … UCX, no undefined-symbol/version errors).
• Ran examples/backends/trtllm/launch/disagg_same_gpu.sh (Qwen3-0.6B, prefill + decode on one GPU): /v1/chat/completions returned status=success with distinct prefill_worker_id and decode_worker_id (elapsed 190 ms, ttft 131 ms) — the NIXL 1.3.1 prefill→decode KV handoff works over UCX.

Note

The local single-host disagg run required working around a pre-existing, NIXL-independent rc22 issue: disagg_machine_id = int(endpoint.connection_id()) % 1021 (components/src/dynamo/trtllm/workers/llm_worker.py:778) can exceed TRT-LLM rc22's NODE_ID_SPACE = 256, raising ValueError: node_id must be in range [0, 256) before the transfer runs. It derives from the runtime connection id (not NIXL), so it is unrelated to this bump. This is already fixed by #12203 (splits the 10-bit machine_id into rc22's 8-bit node_id + 6-bit process_id); once that lands on main, disagg works without any workaround.

---

Summary by CodeRabbit

• New Features
  • Updated the TensorRT-LLM runtime to use NIXL version 1.3.1.
  • Added consistent NIXL version configuration across runtime environments.
• Bug Fixes
  • Improved NIXL installation reliability by validating the configured release and ensuring matching package versions are installed.
  • Added installation verification details to help detect runtime version or library loading issues.

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.

[tanmayv25]

/ok to test ad2a6b5

[coderabbitai]

Walkthrough

The trtllm container configuration now selects NIXL v1.3.1, installs matching NIXL packages from NIXL_REF, validates the release tag, records installation details, and clears NIXL_VERSION consistently across runtime stages.

Changes

NIXL version alignment

Layer / File(s)	Summary
NIXL configuration and dependency declaration container/context.yaml, container/deps/requirements.trtllm.txt	The CUDA 13.1 trtllm default changes to NIXL v1.3.1, and dependency installation is delegated to the runtime Dockerfile.
Runtime installation and environment synchronization container/templates/trtllm_runtime.Dockerfile	The Dockerfile validates NIXL_REF, installs matching NIXL packages, writes version and library-path details, and clears NIXL_VERSION in runtime environment declarations.

Estimated code review effort: 2 (Simple) | ~10 minutes

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The description covers the summary and validation, but it omits required template sections like Details, reviewer start guidance, and Related Issues.	Add the missing template sections, especially Details, Where should the reviewer start?, and a completed Related Issues section or confirmed no-issue checkbox.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly and concisely states the main change: bumping NIXL to v1.3.1 for trtllm.

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

🌿 Fern Docs Preview: https://nvidia-preview-019fab2a-530d-73a8-ad62-a440be423955.docs.buildwithfern.com/dynamo/dev