prompt_models.py

from huggingface_hub import InferenceClient
import re
import os

api_key = f"{os.getenv('ImagiGen_HF_secret')}"


def clean_generated_text(text):
    # Remove asterisks (e.g., **text** or *text*)
    text = re.sub(r"\*+", "", text)

    # Remove special characters (except common punctuation and alphanumeric)
    text = re.sub(r'[^a-zA-Z0-9 .,!?\'"-]', "", text)

    # Normalize multiple spaces into a single space
    text = re.sub(r"\s+", " ", text).strip()
    return text


def generate_prompt_response(api_key, model_name, user_message, max_tokens=1000):
    client = InferenceClient(api_key=api_key)
    messages = [{"role": "user", "content": user_message}]

    # Generate the completion response
    stream = client.chat.completions.create(
        model=model_name, messages=messages, max_tokens=max_tokens, stream=True
    )

    # Collect the response
    response = ""
    for chunk in stream:
        response += chunk.choices[0].delta.content
    return clean_generated_text(response)


def Qwen_72b(user_input):
    model_name = "Qwen/Qwen2.5-72B-Instruct"
    response = generate_prompt_response(api_key, model_name, user_message=user_input)
    return clean_generated_text(response)


def Mixtral(user_input):
    model_name = "mistralai/Mixtral-8x7B-Instruct-v0.1"
    response = generate_prompt_response(api_key, model_name, user_message=user_input)
    return clean_generated_text(response)


def microsoft_phi(user_input):
    model_name = "microsoft/Phi-3-mini-4k-instruct"
    response = generate_prompt_response(api_key, model_name, user_message=user_input)
    return clean_generated_text(response)