[GG] Measure repeatable activation peak after kernel warmup

[voipmonitor]

Problem

determine_available_memory() profiles a warmup pass and a second serving pass inside one peak-memory window. The first pass can include one-time compiler or loader temporaries. A transient on one rank then permanently reduces the distributed KV-cache allocation even though that memory is freed before serving.

This was reproduced with the TP4/DCP4 EXL3 path: identical cold starts varied from 834,560 down to 778,496 KV tokens, with one rank reporting 2.10 GiB peak activation while all ranks had identical live memory before the second pass.

Fix

Reset only the accelerator peak counter after persistent kernel warmup and before the second profile pass. Live allocations are not freed or hidden; persistent buffers remain counted relative to the original baseline. The second pass establishes the repeatable serving activation peak.

A regression test locks the ordering: initial profile, compressor warmup, kernel warmup, peak reset, measured profile.

Validation

• Targeted worker tests: 3 passed
• TP4/DCP4/MTP0, three complete cold starts: 834,560 KV tokens every time; 1.62 GiB activation peak on every rank
• MTP0 decode: 48.00 and 48.58 tok/s, no errors
• TP4/DCP4/MTP3: 485,888 KV tokens; 101.02 tok/s; acceptance 0.6847; no errors
• git diff --check and py_compile: clean

Paired EXL3 consumer: #190

[coderabbitai]

Warning

Review limit reached

@voipmonitor, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 56 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 7597373a-f0cd-48af-b749-4fd279ea8df4

📥 Commits

Reviewing files that changed from the base of the PR and between f978d00 and 703af3f.

📒 Files selected for processing (2)

• tests/v1/worker/test_gpu_worker.py
• vllm/v1/worker/gpu_worker.py

---

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.}

[voipmonitor]

@coderabbitai review

[coderabbitai]

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.