Clay (ryan-chang-33dcc0): Kimi K2.6 rate-limit investigation - #7
Draft
tarunipaleru wants to merge 2 commits into
Draft
Clay (ryan-chang-33dcc0): Kimi K2.6 rate-limit investigation#7tarunipaleru wants to merge 2 commits into
tarunipaleru wants to merge 2 commits into
Conversation
Characterize adaptive TPM headers and 429 behavior for accounts/fireworks/models/kimi-k2p6 across phased, stress, and burst test scenarios. Co-authored-by: tarunipaleru <tarunipaleru@users.noreply.github.com>
Prometheus shows 62 kimi-k2p6 HTTP 429s with rejection_reason=tokens_prompt during a sharp ramp at 2026-08-05 01:21-01:23 UTC, despite smoothed TPM utilization staying well below adaptive ceilings. Co-authored-by: tarunipaleru <tarunipaleru@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Prometheus investigation of why account
ryan-chang-33dcc0(Clay) saw rate limits in the last 24h despite not appearing to hit TPM ceilings on dashboards.Verdict
Clay was rate-limited on kimi-k2p6, but not from sitting at adaptive TPM ceilings.
rejection_reason=tokens_prompton deploymentkimi-k2p6Smoothed utilization vs global RLM limits stayed low (peak ~41% generated, ~13% prompt, ~11–14% cache-adjusted), which is why dashboards look “under limit.” The
tokens_promptrate mirror was also missing/0 during the exact 429 window, so that dimension’s utilization is not visible at rejection time.Artifacts
ryan-chang-33dcc0-rate-limit-investigation.md— full write-up + PromQL