fix(docs): publish Muse Glimmer verification catalog

[yaoyu-33]

Summary

• map the Muse Glimmer verification card to a dedicated canonical model guide
• regenerate the static catalog and Sphinx/Fern model directories
• add an explicit regression assertion for the Muse guide path

Root cause

PR #5479 added the Muse Glimmer model verification card but did not add its canonical guide mapping or regenerate the catalog outputs. The generator therefore fell back to models/README.md. Main run 33719161485 failed test_models_map_to_canonical_guides on core shard 1 and test_generated_outputs_and_navigation_are_current on core shard 0.

Testing

• uv run --no-project --with pytest --with pyyaml python -m pytest --confcutdir=tests/unit_tests/doc_consistency tests/unit_tests/doc_consistency/test_model_verification_catalog.py -q
• uv run --no-project --with pyyaml python skills/create-model-verification-card/scripts/validate_card.py examples/model_verification_cards/muse-glimmer-30b/card.yaml
• uv run --no-project --with pyyaml python scripts/docs/generate_model_verification_catalog.py --check
• uv run --no-project --with pre-commit pre-commit run --all-files

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[yaoyu-33]

/ok to test 26ae2df

[claude]

LGTM — cleanly generated model-verification catalog publication for the Muse Glimmer 30B card; everything is internally consistent.

Verified during review:

• Inventory counts agree across all generated files: 25 model cards / 358 configurations (was 24/345), i.e. +1 card and +13 entries, exactly the 13 Muse Glimmer entries in catalog-v1.json.
• Generator additions line up: the meta-models -> Meta owner label and the muse-glimmer-30b -> models/muse_glimmer/muse-glimmer.md page-path mapping match the source card and the new test assertion.
• Every recipe named in the card commands exists in-tree (muse_glimmer_30b_pretrain_32gpu_h100_bf16_multimodal_config, sft_32gpu_h100_bf16_config, long_context_config, peft_8gpu_h100_bf16_config, and the perf pretrain_32gpu_h100_fp8cs_config).
• README hrefs follow the established convention (fern without .html, sphinx with .html); the Partial status correctly reflects the single unverified item (checkpoint_resume).
• Regeneration drift is already guarded by test_generated_outputs_and_navigation_are_current (runs generate check=True), so stale output would fail CI.

Minor note (not a defect): the Benchmark leaf records precision bf16 while the recipe is fp8cs. Defensible — model/gradient state is BF16, FP8 applies only to GEMMs, and fp8_cs is not an allowed precision enum value (bf16/fp8_mx/nvfp4).

Suggested test cases:

• tests/unit_tests/doc_consistency/test_model_verification_catalog.py::test_models_map_to_canonical_guides
• tests/unit_tests/doc_consistency/test_model_verification_catalog.py::test_catalog_discovers_every_card_without_a_model_list
• tests/unit_tests/doc_consistency/test_model_verification_catalog.py::test_generated_outputs_and_navigation_are_current
• tests/unit_tests/doc_consistency/test_model_verification_catalog.py::test_catalog_is_a_simple_model_directory

No perf tests impacted (no files under scripts/performance/configs/ are touched by this PR).