What's Changed
PR: #952
Added a Cerebras LLM provider for text generation and structured output, and set it as the preferred provider in search operations for faster responses.
-
New Features
- New CerebrasProvider with generate and structured_output; embeddings and rerank are not supported.
- Updated structured output models to be Cerebras-compatible:
- Use fixed-length tuples for keywords and expansions; extra="forbid" on models.
- Introduced ValueMatch, AnyMatch, and RangeSpec for stricter filter validation.
- Normalize JSON schema (remove min/max items; set items:false with prefixItems).
- Added cerebras model config in defaults.yml and moved Cerebras to the top of operation_preferences.
- Factory now detects CEREBRAS_API_KEY and initializes CerebrasProvider.
-
Dependencies
- Added cerebras-cloud-sdk.