handler.py

import torch
import re

from typing import Any, Dict
from transformers import AutoModelForCausalLM, AutoTokenizer


class EndpointHandler:

    def __init__(self, path=""):
        # load model and tokenizer from path
        self.tokenizer = AutoTokenizer.from_pretrained(path)
        self.model = AutoModelForCausalLM.from_pretrained(
            path, device_map="auto", torch_dtype=torch.float16, trust_remote_code=True
        )
        self.device = "cuda" if torch.cuda.is_available() else "cpu"

    def __call__(self, data: Dict[str, Any]) -> Dict[str, str]:
        # process input
        inputs = data.pop("inputs", data)
        parameters = data.pop("parameters", None)
        
        prompt = f'''Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n'''
        prompt += f'### Instruction:\n'
        prompt += f'''Three telecommunications experts analyze the behavior of a phone number based on its call records to evaluate whether it is suspicious behavior. The three experts will discuss together and provide the most confident probability value (0%–100%) of "Yes". No explanation is required.\n\n'''
        prompt += f"### Question:\n{inputs}"
        prompt += f"### Response:\n"

        self.tokenizer.pad_token = self.tokenizer.eos_token
        generation_config = {
            'max_new_tokens':128,
            'top_p':None,
            'do_sample':False,
            'num_beams': 5,
            'temperature':None
        }

        self.model.generation_config.pad_token_id = self.tokenizer.pad_token_id

        llama3_prompt = [
            {
                "role": "system",
                "content": "Below is an instruction that describes a task. Write a response that appropriately completes the request."
            },
            {
                "role": "user",
                "content": ""
            }
        ]
        
        # preprocess
        # inputs = self.tokenizer(prompt, return_tensors="pt").to(self.device)

        llama3_prompt[1]['content'] = prompt
        input_ids = self.tokenizer.apply_chat_template(llama3_prompt, add_generation_prompt=True, add_special_tokens=True, return_tensors="pt").to(self.device)
        prompt_len = input_ids.shape[-1]
        outputs = self.model.generate(input_ids, **generation_config)
        generated_answer = self.tokenizer.decode(outputs[0, prompt_len:], skip_special_tokens=True)
    
        p = self.getProbability(generated_answer)
        if p > 50.0:
            generated_answer = 'Yes'
        else:
            generated_answer = 'No'

        prediction =[{"generated_text": generated_answer}]
        print("--prediction--",prediction )
        return prediction

    def getProbability(text):
        match = re.search(r'(\d+(?:\.\d+)?)%', text)
        if match:
            number = float(match.group(1))
        else:
            # 如果沒有 '%'，提取第一個數字
            match = re.search(r'(\d+(?:\.\d+)?)', text)
            if match:
                number = float(match.group(1))
            else:
                number = None  # 未找到數字

        if number is not None:
            print(number)
            return number
        else:
            # 如果沒有數字，找到第一個 'yes' 或 'no'，不區分大小寫
            match = re.search(r'\b(yes|no)\b', text, re.IGNORECASE)
            if match:
                # 輸出匹配到的 'yes' 或 'no'，保持原始大小寫
                print(match.group(0))
                if match.group(0).lower() == 'yes':
                    return 100.0
                else:
                    return 0.0
            else:
                return 0.0