handler.py

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


class EndpointHandler:
    def __init__(self, path=""):
        # load model and processor from path
        self.pipe = pipeline("text-generation", model="HuggingFaceH4/zephyr-7b-beta", torch_dtype=torch.bfloat16, device_map="auto")

    def __call__(self, data: Dict[str, Any]) -> List[List[Dict[str, str]]]:
        """
        Args:
            data (:dict:):
                The payload with the text prompt and generation parameters.
        """
        # process input
        inputs = data.pop("inputs", data)
        parameters = data.pop("parameters", None)

        if isinstance(inputs, list) and isinstance(inputs[0], list) or isinstance(inputs[0], dict):
            if isinstance(inputs[0], dict):
                inputs = [inputs]
            messages = inputs
        
        else:
            if isinstance(inputs, str):
                messages = [[
                    {
                        "role": "system",
                        "content": "You are a helpful AI assistant",
                    },
                    {"role": "user", "content": inputs},
                ]]
            else:
                messages = [[
                    {
                        "role": "system",
                        "content": "You are a helpful AI assistant",
                    },
                    {"role": "user", "content": input},
                ] for input in inputs]                

        prompts = []
        for message in messages:
            prompts += [self.pipe.tokenizer.apply_chat_template(message, tokenize=False, add_generation_prompt=True)]

        # pass inputs with all kwargs in data
        if parameters is not None:
            # Make endpoint compatible with hf client library
            exclude_list = ["stop", "watermark", "details", "decoder_input_details"]
            parameters = {name: val for name, val in parameters.items() if name not in exclude_list}

            outputs = self.pipe(
                prompts,
                **parameters)
        else:
            outputs = self.pipe(
                prompts, max_new_tokens=32,
            )

        return [{"generated_text": outputs}]