fix(qwen): default PTO2_RING_HEAP to 512MB so a plain Qwen3-14B launch doesn't deadlock

[lterrac]

Summary

A plain pypto-serving ... --model Qwen3-14B (the exact command in the README)
fails to start: the warmup prefill at the default --max-num-batched-tokens 4096 overflows the PTO2 ring-task heap, which defaults to 256MB in the runtime,
and startup aborts with HEAP_RING_DEADLOCK (surfaced as
ACL_ERROR_RT_AICPU_EXCEPTION / code 507018).

This sets a working default so Qwen3-14B launches out of the box, and documents
the knob.

Change

• Qwen314BPyptoExecutor.__init__ does os.environ.setdefault("PTO2_RING_HEAP", 512MB) before any worker is created. It's a setdefault, so an explicit
  PTO2_RING_HEAP from the caller still wins; it's in the executor, so both the
  serving and offline (npu_generate.py) paths get it.
• README: note that Qwen3-14B raises the ring heap to 512MB automatically and
  how to override it.

512MB is the smallest size that clears the warmup on a 14B (256MB deadlocks).

Verified

Launched Qwen3-14B without any PTO2_RING_HEAP in the environment:

• startup reaches Application startup complete, no HEAP_RING_DEADLOCK
• memory breakdown shows the arena at 2.28 GB (512MB x 4 + overhead), confirming
  the 512MB default took effect (256MB would be ~1.1 GB)
• "Once upon a time" → ", there was a young girl named Lily"

Notes

• Qwen3-14B only (the executor is model-specific); DeepSeek path untouched.
• Independent of the staging / kernel-cache PRs.

[coderabbitai]

Important

Review skipped

Auto incremental reviews are disabled on this repository.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 83a3f140-04ec-4702-b88f-1dad9e963a64

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

📝 Walkthrough

Walkthrough

Qwen3-14B now defaults PTO2_RING_HEAP to 512MB before worker creation, preserves existing environment overrides, and documents the runtime behavior and manual override option.

Changes

Qwen3-14B runtime heap

Layer / File(s)	Summary
Configure and document the Qwen3-14B heap default pypto_serving/model/qwen/npu_executor.py, README.md	Defines a 512MB heap default, applies it without overriding an existing environment value, and documents the warmup behavior and PTO2_RING_HEAP override.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Poem

A rabbit saw the heap grow wide,
Five hundred twelve megabytes beside.
Warmup hops without a fright,
Deadlocks vanish out of sight.
Set your own size if you decide!

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the Qwen3-14B heap default change and the deadlock fix.
Description check	✅ Passed	The description accurately describes the heap default, override behavior, and README update.
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.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[gemini-code-assist]

Code Review

This pull request automatically configures the PTO2_RING_HEAP environment variable to 512MB for Qwen3-14B to prevent HEAP_RING_DEADLOCK during warmup, and documents this behavior in the README. The feedback suggests setting this environment variable before calling super().__init__ in the constructor to ensure that any initialization logic in the base class correctly inherits the setting.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.