handler.py

from typing import Any, List, Dict
import torch

from chronos import ChronosPipeline


class EndpointHandler:
    def __init__(self) -> None:
        self.pipeline = ChronosPipeline.from_pretrained("amazon/chronos-t5-tiny")

    def __call__(self, data: Any) -> List[Dict[str, float]]:
        inputs = data.pop("inputs")
        # parameters = data.pop("parameters", {"prediction_length"})

        forecast = self.pipeline.predict(
            torch.tensor(inputs["context"]), prediction_length=5
        )
        return {"forecast": forecast.tolist()}