src/use_llm.py

from huggingface_hub import InferenceClient
import os

# Use a Hugging Face inference endpoint like "google/gemma-1.1-7b-it"
# You must have access to this model (either public or via token)

HUGGINGFACE_API_TOKEN = os.getenv("HF_TOKEN")  # Add this in your HF Space's secret settings
DEFAULT_MODEL = "google/gemma-1.1-7b-it"

client = InferenceClient(DEFAULT_MODEL, token=HUGGINGFACE_API_TOKEN)

def send_chat_prompt(prompt: str, model: str, system_prompt: str) -> str:
    full_prompt = f"<|start_of_turn|>system\n{system_prompt}<|end_of_turn|>\n" \
                  f"<|start_of_turn|>user\n{prompt}<|end_of_turn|>\n" \
                  f"<|start_of_turn|>assistant\n"
    
    response = client.text_generation(
        prompt=full_prompt,
        max_new_tokens=500,
        temperature=0.5,
        stop_sequences=["<|end_of_turn|>"]
    )
    
    return response.strip()

def main_generate(prompt, model=DEFAULT_MODEL, system_prompt="You are a helpful assistant that generates SPARQL queries."):
    response = send_chat_prompt(prompt, model, system_prompt)
    response = response.replace('```', '').replace('json', '').strip()
    return response



HF_TOKEN = os.getenv("HF_TOKEN")
MODEL_ID = "thenlper/gte-large"  # embedding model

client = InferenceClient(model=MODEL_ID, token=HF_TOKEN)

def get_embeddings(texts):
    if isinstance(texts, str):
        texts = [texts]
    embeddings = []
    for text in texts:
        response = client.text_to_vector(text)
        # response is usually a list of floats (the embedding vector)
        embeddings.append(response)
    return embeddings