handler.py

from typing import Dict, List, Any
import json
import numpy as np
from transformers import AutoProcessor, MusicgenForConditionalGeneration
import torch
import logging


class EndpointHandler:
    def __init__(self, path=""):
        # load model and processor from path
        self.processor = AutoProcessor.from_pretrained(path)
        
        # Check if CUDA is available, and set the device accordingly
        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        
        # Load the model to the device
        self.model = MusicgenForConditionalGeneration.from_pretrained(path)
        self.model.to(self.device)  # Correcting this line

    def __call__(self, data: Dict[str, Any]) -> Dict[str, str]:
        """
        Args:
            data (:dict:):
                The payload with the text prompt and generation parameters.
        """
        # Set up logging
        logging.basicConfig(level=logging.DEBUG)
        logger = logging.getLogger(__name__)
        
        # process input
        logger.debug(f"Data: {data}")
        inputs = data.pop("inputs", data)
        logger.debug(f"Inputs: {inputs}")
        parameters = data.pop("parameters", None)
        logger.debug(f"Parameters: {parameters}")
        duration = parameters.pop("duration", None)
        logger.debug(f"Duration: {duration}")
        audio = parameters.pop("audio", None)
        logger.debug(f"Audio: {audio}")
        sampling_rate = parameters.pop("sampling_rate", None)
        logger.debug(f"Sampling Rate: {sampling_rate}")
        if not sampling_rate:
            sampling_rate = self.model.config.audio_encoder.sampling_rate

        if audio is not None:
            audio_array = np.array(audio)
            audio = audio_array[: len(audio_array) // 3]
            # sample["array"] = sample["array"][: len(sample["array"]) // 3]

        if duration is not None:
            # Calculate max new tokens based on duration, this is a placeholder, replace with actual logic
            max_new_tokens = int(duration * 50)  
        else:
            max_new_tokens = 256  # Default value if duration is not provided

        # preprocess
        inputs = self.processor(
            text=[inputs],
            padding=True,
            return_tensors="pt",
            audio=audio,
            sampling_rate=sampling_rate).to(self.device)

        # If 'duration' is inside 'parameters', remove it
        if parameters is not None and 'duration' in parameters:
            parameters.pop('duration')
        if parameters is not None and 'audio' in parameters:
            parameters.pop('audio')
        if parameters is not None and 'sampling_rate' in parameters:
            parameters.pop('sampling_rate')

        # pass inputs with all kwargs in data
        if parameters is not None:
            outputs = self.model.generate(**inputs, max_new_tokens=max_new_tokens, **parameters)
        else:
            outputs = self.model.generate(**inputs, max_new_tokens=max_new_tokens)

        # postprocess the prediction
        prediction = outputs[0].cpu().numpy()

        return [{"generated_text": prediction, "sampling_rate" : sampling_rate}]