[Bizuayeu]

What this changes

zig build test-glm-snapshots-host has not built since #2226: fc8bd41ed has cuda_snaps.zig read
TF_GLM_PROMPT_TAIL through @import("core").profile, and the step's copies module, made in
zig/build/glm_family_host.zig, imports engine_api alone ("no module named 'core' available within module
'copies'"). The module now imports core too, as the step's other modules do. No source file changes. CI does not run
this step, so the break did not show there.

• Issue: n/a (a follow-up to glm cuda: TF_GLM_PROMPT_TAIL=1 keeps raw prompts' tail in the unpaged prompt cache, as the KV pages' store does #2226).
• Stacked on: main 5a45539af; it merges cleanly with 519136dd9, where glm: the config parser reads quantization_config inside text_config too, as EXL3 admission does #2327 has landed since.
• atc overlap --json: at 2026-11-06 22:27 UTC, on this head against main 5a45539af: glm cuda: EXL3 mcg experts at any whole-bit width, mixed by projection #2094 and glm cuda: TF_GLM_GRAPHS=1 replays verify windows as CUDA graphs (#1490) #2279 overlap and
  merge cleanly, no conflicts; incomplete (invalidPulls, otherPullRequestsLimit, pullFilesUnavailable,
  timeLimit).

Checks on this head

• Head commit: 94ec9c4a8
• Lean checker, python3 tools/zig/lean_check.py, on the base and on this head: 155 and 155 problems, none new.
• Tests run, with the command and pass, fail and skip counts: zig build test-glm-snapshots-host 4/4 steps (21 of 21
  tests) on this head; on main 5a45539af it stops at compile with the error above. With CI's flags,
  zig build -Dfixed-cpu test 548/548 steps (7,395 of 7,456 tests, 61 skipped). zig fmt --check build.zig zig.
• CI: all seven checks pass (zig, zig-windows, zig-macos-dev-tools, release-paths, fmt and both
  benchmark-clients).
• Not run, and why: the GPU suites, which this line does not reach.

Receipt

n/a: a build file's module imports for a host test step; it cannot reach inference.

Checklist

• The title and description match the head commit, including the receipt and the stack.
• New tests fail before the change, pass after it, and skip cleanly without their dependency. (No new tests: the
  step's 21 tests build and pass again.)
• Comments and docstrings are one line. No measurements or history in the source.
• No personal data, machine names, internal hosts or local paths in the diff or its commits. No AI attribution lines.

[ashhart]

Thanks @Bizuayeu, a clean find. Main's step can't build because cuda_snaps.zig reads TF_GLM_PROMPT_TAIL through core since #2226, while the copies module only had engine_api. Giving it mods.core, as its sibling module has, is the right fix and touches nothing a build ships. It lands once CI is green, and we'll add the CHANGELOG line at landing.

One optional follow-up, if you'd like it: test-glm-snapshots-host is the one GLM host step not added to the family's aggregate with step.dependOn, which is why CI didn't notice. Adding it would keep it built.

[Bizuayeu]

One optional follow-up, if you'd like it: test-glm-snapshots-host is the one GLM host step not added to the family's
aggregate with step.dependOn, which is why CI didn't notice.

#2403 (comment)

Thanks. Yes, I'll send that as its own pull request once this one lands: the snapshots run joins the family's aggregate
with step.dependOn in glm_family_host.zig, with test-cuda-host run before and after.

[ashhart]

Thank you @Bizuayeu. This is merged into main with your commits under your own name, so you're listed as a TensorFold contributor. It ships in the next release, credited to you in the changelog and the release notes.