handler.py

from typing import Dict, List, Any
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import torch

class EndpointHandler:
    def __init__(self, path=""):
        # Load the model
        tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2-1.5B-Instruct")
        model = AutoModelForCausalLM.from_pretrained(
            "Qwen/Qwen2-1.5B-Instruct",
            torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
            device_map="cuda" if torch.cuda.is_available() else "auto" # Include device_map for correct device allocation
        )

        # Create inference pipeline without specifying the device
        self.pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer)

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

        if isinstance(inputs, str):
            inputs = [inputs]

        # Get predictions from the pipeline
        prediction = self.pipeline(inputs, **parameters)
        
        return prediction

# Example usage
if __name__ == "__main__":
    handler = EndpointHandler()
    data = {
        "inputs": "Hello, how can I",
        "parameters": {"max_length": 50, "num_return_sequences": 1}
    }
    result = handler(data)
    print(result)