Repository navigation
ci(metal): budget the strict gates and cover quantized SET_ROWS - #67
Merged
lalalune merged 1 commit intoOct 14, 2026
Merged
Conversation
#61 made FLASH_ATTN_EXT and MUL_MAT fail the job on timeout, but the full FLASH_ATTN_EXT suite takes ~8.5 min with -t 1 even on an M4 Max, against a 600 s limit on the slower hosted runner. Raise the broad budget to 1800 s, the custom budget to 900 s and the job limit to 120 min. The custom step re-ran the whole MUL_MAT and CPY suites. Filter GET_ROWS, CPY, MUL_MAT and the new SET_ROWS run to the Eliza quant types, so #34's TBQ3_0/TBQ4_0/Q4_POLAR SET_ROWS kernels are now gated on Metal. Update the stale comments that said these kernels were absent. 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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Follow-up from the post-merge review of #61 and #34.
-t 1) the FLASH_ATTN_EXT suite alone takes ~8.5 min against the 600 s limit, so the hosted macOS runner would time out. Broad budget 600→1800 s, custom 300→900 s, job 60→120 min.-pto the Eliza quant types, and SET_ROWS is added (including in the ledger's expected invocations). As a result, metal: custom KV set_rows/cpy for TBQ3_0/TBQ4_0/Q4_POLAR; device-gated FA dequant #34's TBQ3_0/TBQ4_0/Q4_POLAR SET_ROWS kernels are gated on Metal.Verified locally by running this step's script under macOS
/bin/bash3.2 against a Metal build of the current tip (3f131ab): every op passed the verifier; SET_ROWS 12/12, CPY 6/6, MUL_MAT 3/3, ATTN_SCORE_TBQ 6/6, ATTN_SCORE_POLAR 3/3, ISTFT 4/4, GET_ROWS 12 explicitly unsupported, no FAIL lines. Each filtered run takes 1–3 s.🤖 Generated with Claude Code