scorebook.inference_pipeline

Inference pipeline implementation for processing items through model inference.

This module provides a pipeline structure for handling model inference tasks, supporting preprocessing, model inference, and postprocessing steps in a configurable way.

InferencePipeline Objects

class InferencePipeline()

A pipeline for processing items through model inference.

This class implements a three-stage pipeline that handles:

1. Preprocessing of input items
2. Model inference
3. Postprocessing of model outputs

Attributes:

• model - Name or identifier of the model being used
• preprocessor - Function to prepare items for model inference
• inference_function - Function that performs the actual model inference
• postprocessor - Function to process the model outputs

__init__

def __init__(model: str,
             inference_function: Callable,
             preprocessor: Optional[Callable] = None,
             postprocessor: Optional[Callable] = None) -> None

Initialize the inference pipeline.

Arguments:

• model - Name or identifier of the model to use
• inference_function - Function that performs model inference
• preprocessor - Optional function to prepare items for inference.
• postprocessor - Optional function to process model outputs.

run

async def run(items: List[Dict[str, Any]],
              **hyperparameters: Any) -> List[Any]

Execute the complete inference pipeline on a list of items.

Arguments:

• items - List of items to process through the pipeline
• **hyperparameters - Model-specific parameters for inference

Returns:

List of processed outputs after running through the complete pipeline

__call__

async def __call__(items: List[Dict[str, Any]],
                   **hyperparameters: Any) -> List[Any]

Make the pipeline instance callable by wrapping the run method.

Arguments:

• items - List of items to process through the pipeline
• **hyperparameters - Model-specific parameters for inference

Returns:

List of processed outputs after running through the complete pipeline