[danielhanchen]

Repins Inkling (ggml-org#25731) from b49c8ba to 48ad1d8.

The b11871 prebuilt run (40213734709) failed on CPU / linux/arm64 + Vulkan:

ggml-vulkan.cpp:6018:13: error: enumeration value 'GGML_PREC_F32_PEDANTIC' not handled in switch [-Werror,-Wswitch]

Inkling adds GGML_PREC_F32_PEDANTIC and handled it in ggml_vk_prec_allows_f16acc, but not in ggml_vk_src1_prec_allows_q8. 48ad1d8 adds it to the return false group (an F32-pedantic src1 must not be quantized to Q8), which is the only change.

Checks:

• The two switches compiled standalone with -Wswitch -Werror: the old head fails with the same error, the new head passes. No other switch over ggml_prec in the tree lacks the new value.
• llama_pinset.py sim --pin b49c8ba=48ad1d8: b11871 and b11873 merge clean, pin_contract and merge_checks pass, and the merged tree carries the new case.
• Vulkan-only change: the CPU and CUDA builds do not compile this file, so Inkling numerics on those backends are unchanged.

Not covered here: b11887 (newest) refuses #61, #264, #298 and #304 on ggml-cuda/mmq.cu*, which is independent of this repin and needs its own fix before the nightly ages onto it.

[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 2026-10-25T19:26:16.078402Z	8607be2	PR opened
🔒 Security Review	✅ Completed 2026-10-25T19:28:43.554943Z	8607be2	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.