serving: a release carries its own attest wall-time budget; the 27B gets 2.0x - #1761
Merged
Conversation
…ets 2.0x On mainnet 2026-09-08 our 27B card under 20+ concurrent gateway requests ran the attest GEMM chain at 1264-1280 ms against the reference's 740 (1.7x) and failed the 1.6x constant on about one round in seven while serving thousands of tokens - a failed attest zeroes that round's pay. Idle it reads 720-795 ms. The MoE's decode left the card idle enough for the chain to run at speed; a dense model saturates it. `attest.budget_ratio` on the release overrides SERVING_ATTEST_BUDGET_RATIO in judge(); the 27B release carries 2.0 with the measurement noted. Claude-Session: https://claude.ai/code/session_01VjhStJbNW6WzxucTcwuw4y
…dget # Conflicts: # gittensor/validator/weights/serving_loadout.json
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
On mainnet 2026-09-08 our 27B card (UID 111) under 20+ concurrent gateway requests ran the attest GEMM chain at 1264–1280 ms against the reference's 740 ms (1.7×) and failed the 1.6× constant on ~1 round in 7 (19 of 127) while serving — and a failed attest zeroes that round's pay (one such round had 6,196 output tokens). Idle it reads 720–795 ms. The MoE's decode left the card idle enough for the chain to run at speed; a dense model saturates it.
attest.budget_ratioon the release now overridesSERVING_ATTEST_BUDGET_RATIOinjudge(); the 27B release carries 2.0 with the measurement noted in the loadout. Constant unchanged for releases without one. Test added; 806 passed, lint clean.Touches
serving_loadout.json'sattestblock only, so it merges beside #1760 (which edits theauditblock).https://claude.ai/code/session_01VjhStJbNW6WzxucTcwuw4y