fix(gallery): normalize inconsistent tag casing/plurals across gallery models

[Anai-Guo]

Summary

Fixes #9247 — the gallery's Embedding filter only surfaces 5 models because well-known embedding models (nomic-embed-text-v1.5, granite-embedding-*, embeddinggemma-300m, all-MiniLM-L6-v2, bert-embeddings) are tagged with the plural embeddings while the WebUI filter looks for the singular embedding.

While auditing for the reporter's stated suspicion that "this may be happening with other tags", I also found case-inconsistent variants of cpu/gpu/tts/asr (e.g. GPU, Cpu, CPU, TTS, ASR) and three pre-existing duplicate-tag entries.

Changes (only gallery/index.yaml)

Bad tag	Canonical	Models affected
embeddings	embedding	6
TTS	tts	5
ASR	asr	2
CPU, Cpu	cpu	17
GPU	gpu	17

The canonical form is the one already used by 600–800+ existing models AND, for embedding, the term hardcoded by the WebUI category filter:

https://github.com/mudler/LocalAI/blob/master/core/http/views/models.html#L… — { term: 'embedding', label: 'Embedding', icon: 'fas fa-vector-square' }

Bonus cleanup: deduplicated tag entries on 3 models that already had repeated tags before this PR (gpt-oss-20b had gguf listed twice, arcee-ai/AFM-4.5B had gpu listed twice, one Qwen variant had default listed twice).

Verification

$ python -c "import yaml,collections; \
   data = yaml.safe_load(open('gallery/index.yaml')); \
   c = collections.Counter(t for m in data for t in (m.get('tags') or [])); \
   print({k: c[k] for k in ['embedding','embeddings','cpu','CPU','Cpu','gpu','GPU','tts','TTS','asr','ASR']})"
{'embedding': 11, 'embeddings': 0, 'cpu': 806, 'CPU': 0, 'Cpu': 0, 'gpu': 681, 'GPU': 0, 'tts': 166, 'TTS': 0, 'asr': 9, 'ASR': 0}

All 969 model entries unchanged outside their tags lists. Repro for the original bug:

1. Filter by Embedding in the gallery — was 5 models, now 11 (incl. nomic-embed-text-v1.5, granite-embedding-*, embeddinggemma-300m).

🤖 Generated with Claude Code