handler.py

from typing import Any, Dict, List

import torch
import transformers
from transformers import AutoModelForCausalLM, AutoTokenizer


dtype = torch.bfloat16 if torch.cuda.get_device_capability()[0] == 8 else torch.float16

class EndpointHandler:
    def __init__(self, path=""):
        self.tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code=True)
        self.model = AutoModelForCausalLM.from_pretrained(
            path,
            return_dict=True,
            device_map="auto",
            load_in_8bit=True,
            torch_dtype=dtype,
            trust_remote_code=True,
        )
               
        generation_config = self.model.generation_config
        generation_config.max_new_tokens = 200
        generation_config.temperature = 0.8
        generation_config.top_p = 0.8
        generation_config.num_return_sequences = 1
        generation_config.pad_token_id = self.tokenizer.eos_token_id
        generation_config.eos_token_id = self.tokenizer.eos_token_id
        generation_config.early_stopping = True
        self.generate_config = generation_config
        
        self.pipeline = transformers.pipeline(
            "text-generation", model=self.model, tokenizer=self.tokenizer
        )


    def _ensure_token_limit(self, text):
        """Ensure text is within the model's token limit."""
        tokens = self.tokenizer.tokenize(text)
        if len(tokens) > 2048:
            # Remove tokens from the beginning until the text fits
            tokens = tokens[-2048:]
            return self.tokenizer.decode(tokens)
        return text
    
    def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:
        user_prompt = data.pop("inputs", data)
        
        # Add the user's message to the conversation history
        self.conversation_history += f"<user>: {user_prompt}\n"
        
        # Ensure the conversation history is within token limit
        self.conversation_history = self._ensure_token_limit(self.conversation_history)

        # Add the permanent context, user's prompt, and conversation history
        permanent_context = "<context>: You are a life coaching bot with the goal of providing guidance, improving understanding, reducing suffering and improving life. Gain as much understanding of the user before providing guidance."
        structured_prompt = f"{permanent_context}\n{self.conversation_history}<bot> response:"
        
        result = self.pipeline(structured_prompt, generation_config=self.generate_config)
        
        # Extract only the bot's response without the structuring text
        response_text = self._extract_response(result[0]['generated_text'])
        
        # Remove the last "<bot>" from the response_text
        response_text = response_text.rsplit("[END", 1)[0].strip()
        
        # Add the bot's response to the conversation history
        self.conversation_history += f"<bot>: {response_text}\n"
        self.conversation_history = self._ensure_token_limit(self.conversation_history)
    
        return [{"generated_text": response_text}]
        
        return {"response": response_text}