hf_llm.py

# hf_llm.py
from huggingface_hub import InferenceClient
import os

# You can change the default model here:
DEFAULT_MODEL = "mistralai/Mistral-7B-Instruct-v0.2"

# Load token from environment variable for security
HF_API_TOKEN = os.getenv("HF_API_TOKEN", None)

# Client setup
client = InferenceClient(
    model=DEFAULT_MODEL,
    token=HF_API_TOKEN
)

def generate_with_hf(prompt: str, max_new_tokens: int = 256, temperature: float = 0.7) -> str:
    """
    Generate chat-style responses using Hugging Face text generation models.

    Args:
        prompt (str): The instruction or user query.
        max_new_tokens (int): Length of output.
        temperature (float): Controls creativity.

    Returns:
        str: Model response.
    """
    response = client.text_generation(
        prompt,
        max_new_tokens=max_new_tokens,
        temperature=temperature
    )

    # HF returns raw text
    return response