|
| 1 | +from openai import OpenAI |
| 2 | +from LLM.chat import Chat |
| 3 | +from baseHandler import BaseHandler |
| 4 | +from rich.console import Console |
| 5 | +import logging |
| 6 | +import time |
| 7 | +logger = logging.getLogger(__name__) |
| 8 | + |
| 9 | +console = Console() |
| 10 | +from nltk import sent_tokenize |
| 11 | + |
| 12 | +class OpenApiModelHandler(BaseHandler): |
| 13 | + """ |
| 14 | + Handles the language model part. |
| 15 | + """ |
| 16 | + def setup( |
| 17 | + self, |
| 18 | + model_name="deepseek-chat", |
| 19 | + device="cuda", |
| 20 | + gen_kwargs={}, |
| 21 | + base_url =None, |
| 22 | + api_key=None, |
| 23 | + stream=False, |
| 24 | + user_role="user", |
| 25 | + chat_size=1, |
| 26 | + init_chat_role="system", |
| 27 | + init_chat_prompt="You are a helpful AI assistant.", |
| 28 | + ): |
| 29 | + self.model_name = model_name |
| 30 | + self.stream = stream |
| 31 | + self.chat = Chat(chat_size) |
| 32 | + if init_chat_role: |
| 33 | + if not init_chat_prompt: |
| 34 | + raise ValueError( |
| 35 | + "An initial promt needs to be specified when setting init_chat_role." |
| 36 | + ) |
| 37 | + self.chat.init_chat({"role": init_chat_role, "content": init_chat_prompt}) |
| 38 | + self.user_role = user_role |
| 39 | + self.client = OpenAI(api_key=api_key, base_url=base_url) |
| 40 | + self.warmup() |
| 41 | + |
| 42 | + def warmup(self): |
| 43 | + logger.info(f"Warming up {self.__class__.__name__}") |
| 44 | + start = time.time() |
| 45 | + response = self.client.chat.completions.create( |
| 46 | + model=self.model_name, |
| 47 | + messages=[ |
| 48 | + {"role": "system", "content": "You are a helpful assistant"}, |
| 49 | + {"role": "user", "content": "Hello"}, |
| 50 | + ], |
| 51 | + stream=self.stream |
| 52 | + ) |
| 53 | + end = time.time() |
| 54 | + logger.info( |
| 55 | + f"{self.__class__.__name__}: warmed up! time: {(end - start):.3f} s" |
| 56 | + ) |
| 57 | + def process(self, prompt): |
| 58 | + logger.debug("call api language model...") |
| 59 | + self.chat.append({"role": self.user_role, "content": prompt}) |
| 60 | + |
| 61 | + language_code = None |
| 62 | + if isinstance(prompt, tuple): |
| 63 | + prompt, language_code = prompt |
| 64 | + |
| 65 | + response = self.client.chat.completions.create( |
| 66 | + model=self.model_name, |
| 67 | + messages=[ |
| 68 | + {"role": self.user_role, "content": prompt}, |
| 69 | + ], |
| 70 | + stream=self.stream |
| 71 | + ) |
| 72 | + if self.stream: |
| 73 | + generated_text, printable_text = "", "" |
| 74 | + for chunk in response: |
| 75 | + new_text = chunk.choices[0].delta.content or "" |
| 76 | + generated_text += new_text |
| 77 | + printable_text += new_text |
| 78 | + sentences = sent_tokenize(printable_text) |
| 79 | + if len(sentences) > 1: |
| 80 | + yield sentences[0], language_code |
| 81 | + printable_text = new_text |
| 82 | + self.chat.append({"role": "assistant", "content": generated_text}) |
| 83 | + # don't forget last sentence |
| 84 | + yield printable_text, language_code |
| 85 | + else: |
| 86 | + generated_text = response.choices[0].message.content |
| 87 | + self.chat.append({"role": "assistant", "content": generated_text}) |
| 88 | + yield generated_text, language_code |
| 89 | + |
0 commit comments