flow_inference.model_handling.ModelManager

class flow_inference.model_handling.ModelManager

Load TrOCR models and processors on the best available device.

The manager selects CUDA, Apple MPS, or CPU as the inference device and provides helpers for loading Hugging Face vision-encoder-decoder models and TrOCR processors.

__init__()

Initialize the model manager and select an inference device.

Methods

__init__()	Initialize the model manager and select an inference device.
load_model(model_name)	Load a TrOCR-compatible vision-encoder-decoder model.
load_processor(processor_name)	Load a TrOCR processor with fast-tokenizer fallback.