viettranx 28a17c4d6b feat(embedding): data-driven dimension config with multi-provider support ...

- Add `dimensions` field to EmbeddingSettings for per-provider truncation
- Default to 1536 dims to match pgvector schema, preventing INSERT failures
- Replace hardcoded Gemini-only dimension truncation with data-driven config
- Fix verify endpoint to apply dimensions and return mismatch warning
- Expand curated embedding models: Gemini (gemini-embedding-001), Cohere
  (embed-v4), DashScope (text-embedding-v3), OpenAI (text-embedding-3-large)
- Add dimensions input to provider detail UI with i18n (en/vi/zh)
- Add anthropic_native to NoEmbeddingTypes (incompatible auth format)
- System settings verify always requests 1536d to match production behavior

2026-03-25 14:50:43 +07:00

..

web

feat(embedding): data-driven dimension config with multi-provider support

2026-03-25 14:50:43 +07:00