Repository navigation
vision: GGUF think-on across 0.32.14, 0.33.2, 0.33.3 and main — no build stands out - #285
Merged
Merged
Conversation
…ild stands out Three think-on legs (33.7 h, 54 rung-suites, 0 OOMs). The pair that shares the native payload (0.33.3 vs main) differs on 10/56 quality cells, the same as 0.33.2 vs 0.33.3 (11/56): the think-on run-to-run spread, driven by trace length capping an arm and the ladder re-answering it at a bigger rung. gemma4 is stable across the 0.33 builds; qwen3.6's scene cell is flagged for an n>=5 repeat. 0.33.3's three 1800 s timeouts were daytime contention (main finished the same arms overnight). Generator render appended; sync task 7d extended. Co-Authored-By: Claude Fable 5.1 <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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Appends row 2 (GGUF think-on, three builds, 2026-09-06 22:21 → 09-08 08:02) to the GGUF campaign doc and extends sync task 7d.
mainfinished the same arms overnight.main(same native payload) 10/56 quality cells, 0.33.2 vs 0.33.3 11/56, 0.32.14 vs 0.33.3 19/56; contract 7 / 15 / 16 of 64. The payload-sharing pair moves as much as the others, so this is the think-on run-to-run spread (trace length → cap → ladder escalation → a different answer), not a build effect.scene_singlethink-on (0.972 on 0.33.2 at 16384; 0.238 on 0.33.3 and main after escalation to 32768) is flagged for an n ≥ 5 repeat on both builds in one container before any call. nemotron and qwen3.8 move in no build's favour.render-ggml.sh onoutput appended (T1 ×4, T2 ×8, contract matrices ×4); the derived section names the tables as its veto.🤖 Generated with Claude Code