app.py

from llama_cpp import Llama
from typing import Optional, Dict, Union
from huggingface_hub import hf_hub_download
import gradio as gr
import time



model_path = hf_hub_download(
    repo_id="omeryentur/phi-3-sql",
    filename="phi-3-sql.Q4_K_M.gguf",
    use_auth_token=True
)

llm = Llama(
    model_path=model_path,
    n_ctx=512,
    n_threads=1,
)

def generate_sql_query(text_input_schema:str,text_input_question: str):
    global pattern_counter
    
    try:
        prompt = f"""<|system|>
            {table_info}

            <|user|>
            {question}

            <|sql|>"""
        
        completion = llm(
            prompt,
            max_tokens=512,
            temperature=0,
            stop=["<end_of_turn>"]
        )
        
        generated_pattern = completion['choices'][0]['text'].strip()
        result = parse_log_with_grok(text, generated_pattern)
        time.sleep(0.5)
            
        return result
        
    except Exception as e:
        return {"error": e}

with gr.Blocks() as demo:
    gr.Markdown("# Sql Query")
    
    with gr.Row():
        with gr.Column():
            text_input_schema = gr.TextArea(label="Schema")
            text_input_question = gr.Textbox(label="question")
            generate_btn = gr.Button("Create Sql Query")
        
    with gr.Row():
        with gr.Column():
            output = gr.JSON(label="Sql Query:")
    
    generate_btn.click(
        fn=generate_sql_query,
        inputs=[text_input_schema,text_input_question],
        outputs=[output]
    )

if __name__ == "__main__":
    demo.launch()