OpenAI-compatible structured-output controls. Forwarded to the worker
unchanged; honored natively by llama-server (compiles JSON Schema → GBNF
internally) and vLLM (xgrammar / outlines). Per-request opt-in — when
omitted, the model generates without constraint.
OpenAI-compatible structured-output controls. Forwarded to the worker unchanged; honored natively by llama-server (compiles JSON Schema → GBNF internally) and vLLM (
xgrammar/outlines). Per-request opt-in — when omitted, the model generates without constraint.