docs(fold): the deploy row adds gfx1151's build, measured on the build itself - #413
Merged
Merged
Conversation
…d itself CUDA (#411) and Metal (#412) each added a sentence on their deployed build, verified on the build itself. gfx1151's is its production-configuration baseline on the promoted image, recorded in the gate (#403, #410): think-off 135 of 135 blocks valid, OCRBench 172/200 on the q8_0 run's items, preflight PASS=24 SKIP=8 with rocm7's quality floors and fp16 canary, and think-on under f16. It says "measured", not "verified": under f16 the think-off cells are not equal to the gate's q8_0 cells, and gemma4:26b's think-on leaves three cases unfinished against one. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The fold record's "tag and deploy" row now carries a sentence per host on its deployed build. CUDA's came in #411 and Metal's in #412. This adds gfx1151's, at the maintainer's word, and links it to the gate's 2026-09-28 baseline (#403, #410).
The sentence: in a bench container with production's environment, all 135 think-off blocks of five models finish with valid JSON, and OCRBench scores 172/200 on the same items as the gate's
q8_0run. The preflight passes at PASS=24 SKIP=8, with rocm7's new quality floors and fp16 canary. Under f16, think-on finishes every case on gemma4:31b and qwen3.8, loops on one sampled nemotron3 cell in 54, and leaves three gemma4:26b cases unfinished, against one underq8_0.It says "measured", not "verified". CUDA's and Metal's sentences compare their build with a reference under the same settings. gfx1151's baseline ran under f16, production's KV type since ADR 0043. So its think-off cells are not equal to the gate's
q8_0cells, and gemma4:26b's think-on differs too, as the gate records.Verification
check_source_paths.py --changed-since origin/mainis clean, and the link's anchor exists inamd-upgrade-gate.md.amd-server/rocm-gfx1151🤖 Generated with Claude Code