[danielhanchen]

Pins #304 (CUDA: add IQ1_M to MMQ) at 64b2a7e. Nothing upstream broke; this adds a feature. IQ1_M had no MMQ kernel, so batched IQ1_M matmuls (prompt processing, and MoE MUL_MAT_ID in particular) fell back to dequantize + cuBLAS. No equivalent PR is open on ggml-org/llama.cpp.

• pr-set.json: CUDA: add IQ1_M to MMQ #304 appended after model : DeepSeek-V4.1-Flash-1017 full stack pin (all DeepSeek-V4.1 PRs except #292) #298.
• feature-checks.json: new iq1m-mmq feature owned by CUDA: add IQ1_M to MMQ #304. It runs the IQ1_M MUL_MAT cases with m = 129 and the MUL_MAT_ID cases with 256/512 experts that CUDA: add IQ1_M to MMQ #304 adds. Both filters match 0 cases on a tree without the pin, so the check fails if the pin is missing.

Evidence

llama_pinset.py sim with this pin set:

tag	merge	pin_contract	merge_checks
b11756 (aged base now)	clean, 21 pins	all intact	clean
b11757	clean	all intact	clean
b11759	clean	all intact	clean
b11760, b11761	ggml-org#25731 REFUSED	1 pin not intact	clean

The b11760+ refusal is ggml-org#25731 (Inkling) and is the same on the current master pin set without this change; #304 merges cleanly there too.

llama_pinset.py verify --cuda on b11756 and b11759 (B200):

check	b11756	b11759
CPU gate build + feature_matrix CPU	ok, 13/13	ok, 13/13
full CUDA build	ok	ok
test-chat	pass	pass
test-backend-ops MUL_MAT / MUL_MAT_ID	2582/2582, 1027/1027	2582/2582, 1027/1027
test-backend-ops PAD / GET_ROWS	37/37, 349/349	37/37, 349/349
feature_matrix --gpu	13/13	13/13
new iq1m-mmq check, --gpu	12/12 + 16/16 matched CPU	12/12 + 16/16 matched CPU

audit --pr-set: #304 ok (base base/upstream-d80619397, GitHub diff == the pin's own change, +256/-2 in 16 files); feature-checks coverage ok. The FIX rows for #264, #266 and #298 are the same on the current master pin set.

Kernel correctness, KLD vs the CPU backend and the llama-bench numbers (about 3x prompt processing on an IQ1_M MoE, decode unchanged) are in #304.

Not covered

• ROCm, Vulkan, macOS, Windows and ARM legs only run in the nightly.
• Tags newer than b11761 were not simmed.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-21T00:21:30.308956Z	eb114cf	PR opened
🔒 Security Review	✅ Completed 2026-10-21T00:24:32.906798Z	eb114cf	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.