bio_reasoning.layers.b package#

Layer B consists of bespoke foundation models specialized for non-textual or multimodal data (e.g. genomic sequences, protein structures, images) that interface with the LLM.

bio_reasoning.layers.b.visual_describer_factory(api_key: str, api_base_url: str, model_name: str = 'gpt-4.1', system_prompt: str | None = None) → Callable[[str | List[str], str], str][source]#

Factory function to create a visual description function with provided configuration.

Parameters:

  • api_key (str) – The API key for authentication.

  • api_base_url (str) – The base URL of the API providing the visual description service.

  • model_name (str, optional) – Name of the model. Defaults to ‘gpt-4.1’.

  • system_prompt (str, optional) – A prompt to set the system context. Defaults to None.

Returns:

A function that accepts image URIs and user prompt to generate a visual description.

Return type:

Callable[[Union[str, List[str]], str], str]

bio_reasoning.layers.b.load_image_data(uri: str) → str[source]#

Load image data and return either as base64 encoded string (for local files) or direct URL (for remote resources).

Parameters:

uri – Image URI (file path or URL)

Returns:

Either a base64 encoded string (local files) or direct URL (remote resources)

Submodules#