handler.py

from typing import Any, Dict

import torch
from transformers import AutoModel, AutoProcessor


class EndpointHandler:
    def __init__(self, path=""):
        # load model and processor from path
        self.processor = AutoProcessor.from_pretrained("suno/bark-small")
        self.model = AutoModel.from_pretrained(
            "suno/bark-small",
        ).to("cuda")

    def __call__(self, data: Dict[str, Any]) -> Dict[str, str]:
        """
        Args:
            data (:dict:):
                The payload with the text prompt and generation parameters.
        """
        # process input
        text = data.pop("inputs", data)
        voice_preset = data.get("voice_preset", None)
        if voice_preset:
            inputs = self.processor(
                text=[text],
                return_tensors="pt",
                voice_preset=voice_preset,
            ).to("cuda")
        else:
            inputs = self.processor(
                text=[text],
                return_tensors="pt",
            ).to("cuda")

        with torch.autocast("cuda"):
            outputs = self.model.generate(**inputs)

        # postprocess the prediction
        prediction = outputs.cpu().numpy().tolist()

        return {"generated_audio": prediction}