ModelInfo

ModelInfo#

class rapidtools.models.ModelInfo(key: str, label: str, kind: str, description: str, module: str, class_name: str, default_model: str, catalog: list[dict[str, str]]=<factory>, attribute_prefix: str = '', key_env: tuple[str, ...]=(), key_label: str = '', analyzer_default_model: str | None = None, supports_4bit: bool = False, supports_batch: bool = False, free_form_model: bool = False)[source]#

Bases: object

Static description of a model backend.

Variables:

  • key (str) – Short registry key ('gemini', 'qwen_vl', …).

  • label (str) – Human-readable provider name.

  • kind (str) – 'api' (hosted, needs a key), 'local' (Hugging Face weights) or 'segmentation' (SAM 3).

  • description (str) – One-line description shown in user interfaces.

  • module (str) – Import path of the module holding the wrapper class.

  • class_name (str) – Name of the wrapper class inside module.

  • default_model (str) – Model ID used when none is given to the wrapper.

  • catalog (list[dict[str, str]]) – Curated {'model_id', 'family', ...} records.

  • attribute_prefix (str) – Prefix used for attributes written by analyzers.

  • key_env (tuple[str, ...]) – Environment variable(s) holding the API key (API kinds).

  • key_label (str) – Label for the key field in user interfaces.

  • analyzer_default_model (str | None) – Cheaper default used by batch analyzers and the GUI; falls back to default_model.

  • supports_4bit (bool) – Whether load_in_4bit is accepted.

  • supports_batch (bool) – Whether the wrapper offers run_inference_batch.

  • free_form_model (bool) – Whether arbitrary model IDs beyond the catalogue are expected to work (generic Hugging Face backends).

Example

>>> from rapidtools.models.catalogs import PROVIDERS
>>> PROVIDERS['claude'].kind, PROVIDERS['gemma4'].supports_batch
('api', True)
property model_ids: list[str]#

Model IDs of the catalogue, with the analyzer default first.

property is_api: bool#

True for hosted providers that need an API key.