Repin GLM-5-Next (#27754) to the head that builds under fatal warnings

[danielhanchen]

Why

Nightly run 34412786914 (base b10881) did not publish: both macOS legs failed in the build step and assemble refused to publish.

src/models/glm5next.cpp:412:5: error: 'ggml_mul_mat_set_prec' is deprecated: use ggml_prec_set_acc() instead [-Werror,-Wdeprecated-declarations]

Upstream deprecated the precision setters in ggml-org#26675, and the macOS prebuilt legs build with LLAMA_FATAL_WARNINGS=ON under clang. The offending file comes from pin 6, ggml-org#27754 (branch unslothai:glm5next/upstream), which was 252 commits behind upstream master.

What changed on the pin

• 2af9da2d5 merges upstream master into glm5next/upstream (clean, no conflicts).
• d94f44e79 switches the one call at src/models/glm5next.cpp:412 to ggml_prec_set_acc(w, GGML_PREC_F32). Same op_params slot, same F32 accumulation for the indexer weights.

This PR moves scripts/unsloth/pr-set.json entry 6 from b9b8207fc to d94f44e79.

Evidence

• Fatal-warnings build of the merged pin: llama, llama-cli, test-llama-archs build clean; test-llama-archs -a glm5next -s 1234 reports OK (0.00e+00).
• Local replay of the resolve loop on b10881 with the updated pin set: all 13 pins merge (Inkling and mtmd CMakeLists via additive merge as before), merge_checks.py clean, pin_contract.py OK, mix head ee659afe2.
• grep over the mixed tree for ggml_mul_mat_set_prec( and ggml_flash_attn_ext_set_prec( outside the ggml headers: zero call sites left.

After merge

Tonight's schedule picks up master. To recover the b10881 release earlier, re-run the failed jobs of run 34412786914 after this lands.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Repo admins can enable using credits for code reviews in their settings.