impresso_langident_wrapper.py

from transformers import Pipeline
from transformers.pipelines import PIPELINE_REGISTRY
import floret
from huggingface_hub import hf_hub_download



class Pipeline_One(Pipeline):
    # def __init__(self, model_path: str):
    #     """
    #     Initialize the Floret language detection pipeline
        
    #     Args:
    #         model_path (str): Path to the .bin model file
    #     """
    #     super().__init__()
    #     self.model = floret.FastText.load_model(model_path)
    


    def _sanitize_parameters(self, **kwargs):
        # Add any additional parameter handling if necessary
        return {}, {}, {}

    def preprocess(self, text, **kwargs):
        return text

    def _forward(self, inputs):
        model_output = self.model.predict(**inputs, k=1)

        return model_output

    def postprocess(self, outputs, **kwargs):
        return outputs
    


PIPELINE_REGISTRY.register_pipeline(
    task="language-detection",
    pipeline_class=Pipeline_One,
    default={"model": None},
)