multipurpose_chatbot/engines/base_engine.py

import os
import numpy as np
from huggingface_hub import snapshot_download
# ! Avoid importing transformers
# from transformers import AutoConfig, AutoTokenizer, PreTrainedTokenizer
from typing import Any, Callable, Dict, Generator, List, Optional, Tuple, Union
import time


class BaseEngine(object):
 def __init__(self, **kwargs) -> None:
 pass

 @property
 def max_position_embeddings(self) -> int:
 return 10000

 @property
 def tokenizer(self):
 raise NotImplementedError
 
 @property
 def processor(self):
 raise NotImplementedError
 
 def load_model(self, ):
 raise NotImplementedError

 def generate_yield_string(self, prompt, *args, **kwargs):
 raise NotImplementedError
 
 def generate_yield_string_final(self, *args, **kwargs):
 output = None
 for out in self.generate_yield_string(*args, **kwargs):
 output = out
 return output
 
 def generate_yield_string(self, prompt, *args, **kwargs):
 raise NotImplementedError
 
 def generate_yield_string_final(self, *args, **kwargs):
 output = None
 for out in self.generate_yield_string(*args, **kwargs):
 output = out
 return output

 def apply_chat_template(self, conversations, add_generation_prompt: bool, add_special_tokens=False, **kwargs) -> str:
 """
 return string convo, add_special_tokens should be added later
 """
 bos_token = self.tokenizer.bos_token
 eos_token = self.tokenizer.eos_token
 if not add_special_tokens:
 # prevent bos being added to string
 self.tokenizer.bos_token = ""
 self.tokenizer.eos_token = ""
 full_prompt = self.tokenizer.apply_chat_template(
 conversations, add_generation_prompt=add_generation_prompt,
 tokenize=False,
 )
 self.tokenizer.bos_token = bos_token
 self.tokenizer.eos_token = eos_token
 return full_prompt