semantic_router.decision_models.gliner2

GLiNER2 decision models, run locally: fastino/GLiNER2.5-Decide by default.

GLiNER2DecisionModel Objects

class GLiNER2DecisionModel(DecisionModel)

A GLiNER2 decision model run in-process with the gliner2 library.

Answers choice, noul and score questions locally, with no network calls and no key. Requires the local extra: pip install "semantic-router[local]".

Arguments:

• name: The model on the Hugging Face Hub, or a local directory.
• device: "cuda", "mps" or "cpu". Left out, the first available.
• batch_size: States per forward pass when answering many at once.
• classifier: A loaded gliner2.classification.Classifier to use instead of loading name, e.g. one with a fine-tuned adapter.

ask_many

def ask_many(states: Sequence[object],
             questions: dict[str, Question]) -> list[dict[str, Answer]]

answer the same questions about many states in batched forward passes.

ask

def ask(state: object, questions: dict[str, Question]) -> dict[str, Answer]

answer every question about one state in one forward pass.