components/llm/llm_api.py

import os
from threading import Lock
from components.llm.common import LlmParams, LlmPredictParams
from components.llm.deepinfra_api import DeepInfraApi

class LlmApi:
    _instance = None
    _lock = Lock()

    def __new__(cls):
        with cls._lock:
            if cls._instance is None:
                cls._instance = super(LlmApi, cls).__new__(cls)
                cls._instance._initialize()
        return cls._instance

    def _initialize(self):
        LLM_API_URL = os.getenv("LLM_API_URL", "https://api.deepinfra.com")
        LLM_API_KEY = os.getenv("DEEPINFRA_API_KEY", "")
        LLM_NAME = os.getenv("LLM_NAME", "meta-llama/Llama-3.3-70B-Instruct-Turbo")
        TOKENIZER_NAME = os.getenv("TOKENIZER_NAME", "unsloth/Llama-3.3-70B-Instruct")

        default_llm_params = LlmParams(
            url=LLM_API_URL,
            api_key=LLM_API_KEY,
            model=LLM_NAME,
            tokenizer=TOKENIZER_NAME,
            context_length=130000,
            predict_params=LlmPredictParams(
                temperature=0.15, top_p=0.95, min_p=0.05, seed=42,
                repetition_penalty=1.2, presence_penalty=1.1, n_predict=6000
            )
        )
        self.api = DeepInfraApi(default_llm_params)

    def get_api(self):
        return self.api