[inureyes]

Summary

Closes #2445. Five follow-ups left by #2438 and #2439, in one PR as the issue asks, plus one blocker found while verifying item 1 on a real checkpoint.

Changes

1. kv_arch reads Nemotron-H layer_types (src/execution/kv_arch.rs). For model_type nemotron_h without a hybrid_override_pattern, the layers_block_type / layer_types array is mapped through NEMOTRON_H_BLOCK_NAMES (now pub(crate) in src/models/nemotron_h.rs, not copied), * entries are counted, and the layer count is the array length, the way NemotronHConfig::post_init derives it. linear_attention is an M layer and is no longer counted by the contains("attention") filter.
   • Blocker found during verification: estimate_kv_arch parsed config.json with plain serde, and Nemotron-H configs such as nvidia-nemotron-3-nano-30b-a3b-4bit carry time_step_limit: [0.0, Infinity], so the KV estimate was "unavailable" for them, pattern or not (main reports KV cache: 0 bytes (unavailable ...) for that checkpoint). The file now goes through sanitize_config_json first, as the model loader does. Without this, item 1 has no effect on a real checkpoint. This also makes the estimate available for any other config that carries Infinity / NaN, which used to report 0 KV bytes.
2. No stale-index warning on OptiQ checkpoints (src/lib/mlxcel-core/src/weights.rs, src/loading/vlm.rs). optiq_vision_sidecar_path moves into mlxcel_core::weights (same validation: relative, normal components only, .safetensors), and loading::vlm reuses it. collect_shard_paths drops exactly the declared sidecar from the index's shard list before validate_index_shards, and only when it names a subdirectory path (a top-level name is an ordinary shard). The config is read with the same Infinity / NaN replacement the model loader applies. An undeclared subdirectory entry, a malformed sidecar path or an unreadable config leaves the list unchanged, so those still fail validation and take the glob fallback as before.
3. Text-only Muse Glimmer refuses image and video markers (src/multimodal/muse_glimmer_runtime.rs, src/models/muse_glimmer.rs, src/loading/vlm_muse_glimmer.rs, src/server/model_worker.rs, src/multimodal/vlm_runtime.rs). The check is split into a helper over token ids. MuseGlimmerTextWrapper keeps the checkpoint's image_token_id / video_token_id (two i32 fields, set by load_muse_glimmer_text), and reject_muse_glimmer_text_only_markers runs at both sites for LoadedModel::MuseGlimmer. The error says the checkpoint is loaded text-only.
4. Integer LoRA factors no longer abort fusion (src/lora/loader.rs). compute_lora_delta casts both factors to f32 before the matmul, as runtime_lora (at claim) and src/lora/embedding.rs do. The delta was already f32 because the scale is an f32 array, so for a floating-point pair only the matmul's input precision changes; for an f32 adapter (what mlx_lm.lora writes) nothing changes.
5. Nemotron-H surfaces the LoRA width error (src/models/nemotron_h.rs). The shared-expert up_proj / down_proj and the two latent projections return None only when {prefix}.weight is absent; otherwise the constructor error is propagated with ?, so the claim_linear width error from fix(lora): pin runtime LoRA term widths on quantized linear bases #2438 is the load error instead of a later "unclaimed" failure.

Tests

Each new test fails on the reverted fix (each fix reverted in place, test binary rebuilt, then restored):

Test	Reverted result
execution::kv_arch::tests::nemotron_h_layer_types_match_the_equivalent_pattern	panics: no estimate for the layer_types config
execution::kv_arch::tests::a_config_file_with_infinity_still_estimates	panics: no estimate (serde rejects Infinity)
weights::tests::test_collect_shard_paths_skips_only_the_declared_optiq_sidecar (mlxcel-core)	left: ["model.safetensors", "stray.safetensors"], glob fallback taken
loading::vlm::muse_glimmer::tests::text_only_muse_glimmer_refuses_image_and_video_markers	panics: text-only model accepted marker (call sites reverted)
lora::loader::tests::an_integer_lora_pair_fuses_like_the_same_values_in_f32	process abort, exit 134: [matmul] Only inexact types are supported but int32 and int32 were provided
models::nemotron_h::tests::a_wrong_width_lora_term_on_an_optional_projection_fails_the_load	panics: load succeeded without the shared experts

The Nemotron-H test is cfg(not(all(target_os = "macos", feature = "metal"))), since Metal refuses runtime LoRA for this model before any layer is built.

Verification

GB10, --features cuda, test-fast profile, every GPU run under gpu-lock, MLXCEL_SDPA_DETERMINISTIC=1. "Main" is a binary built from 74f9ae1 in the same worktree; "fixed" is this branch.

• cargo test --profile test-fast --features cuda --lib -- kv_arch lora:: muse_glimmer nemotron_h loading::vlm multimodal:: model_worker memory_estimate --test-threads=1: 877 passed, 0 failed.
• cargo test --profile test-fast -p mlxcel-core --features cuda --lib -- weights:: runtime_lora --test-threads=1: 38 passed.
• cargo clippy --profile test-fast --features cuda --lib --tests -- -D warnings, cargo fmt --all -- --check, and the five contract checks (check_llama_compat_manifest, check_kernel_dtype_keys, check_kernel_port_dispatch, check_crate_versions, insert_apache_header --check): clean.

Real checkpoints (~/models_archive/mlx/...):

Item	Checkpoint and command	Main	Fixed
1	nvidia-nemotron-3-nano-30b-a3b-4bit, mlxcel inspect -m	KV cache: 0 bytes (unavailable ...)	48.0 MiB (hybrid: 6 attention layer(s) hold KV)
1	same checkpoint with a transformers 5 style config (hybrid_override_pattern rewritten as layer_types, num_hidden_layers removed), mlxcel inspect -m	0 bytes (unavailable ...)	48.0 MiB (hybrid: 6 attention layer(s) hold KV), identical to the pattern config
2	lfm2.5-vl-3b-optiq-4bit, mlxcel generate, text prompt and --image tests/fixtures/test_image_shapes.png	stale-index warning printed	no warning; generated text identical to main for both prompts
2	muse-glimmer-30b-optiq-4bit, mlxcel generate --image ... --show-reasoning	stale-index warning printed	no warning; generated text identical to main
3	text-only view of muse-glimmer-30b-optiq-4bit (top-level shards, index without the sidecar entries, no optiq_vision; detected as MuseGlimmer), prompt "<|image|> What is in the picture?"	accepted, generates from the marker ids	Error: this Muse Glimmer checkpoint is loaded text-only (no vision tower), but the prompt contains image or video marker tokens ..., exit 1
3	same view, plain prompt	generates	identical text
4	llama-3.2-1b-instruct-bf16 (dense), mlxcel generate --adapter, q_proj adapter of layers 14 and 15 trained with mlx_lm.lora, factors rounded to round(x * 100) and stored as int32, scale divided by 1e4	terminate called after throwing an instance of 'std::invalid_argument' ... [matmul] Only inexact types are supported, exit 134	loads and generates; text identical to the same integer values stored as f32
4	same, f32 twin adapter	generates	identical text to main
5	nvidia-nemotron-3-nano-30b-a3b-4bit (n_shared_experts: 1, so every MoE layer builds its shared experts through the new path), mlxcel generate --show-reasoning, 24 tokens	generates	identical text

The CLI's media preparation goes through prepare_request_vlm_embeddings, the server worker site, so the item 3 run exercises that call site. The unit test covers the prepare_vlm_embeddings_with_cache site and the helper the worker calls; no unit test drives prepare_request_vlm_embeddings itself, so the worker arm is covered only by the real run.

Not in this PR

Review found the same declared-sidecar entry treated as a bad shard in two other places, outside this issue's scope: src/downloader/completeness.rs (missing_indexed_shards counts optiq/optiq_vision.safetensors as missing, so a store-resolved OptiQ snapshot classifies as incomplete) and src/server/webui/catalog_fs.rs (reports it as an unsafe shard path). weight_footprint_bytes also leaves the sidecar's bytes out of the memory preflight. Both can reuse mlxcel_core::weights::optiq_vision_sidecar_path.