app.py

# app.py
# Stable CPU-only Hugging Face Space
# Phi-3-mini + LoRA (NO bitsandbytes, NO SSR issues)

import warnings
warnings.filterwarnings("ignore", category=FutureWarning)

import torch
import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

# ─────────────────────────────────────────────
# Config
# ─────────────────────────────────────────────
BASE_MODEL = "unsloth/Phi-3-mini-4k-instruct"
LORA_PATH  = "saadkhi/SQL_Chat_finetuned_model"

MAX_NEW_TOKENS = 180
TEMPERATURE    = 0.0
DO_SAMPLE      = False

# ─────────────────────────────────────────────
# Load model & tokenizer (CPU SAFE)
# ─────────────────────────────────────────────
print("Loading base model on CPU...")

model = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL,
    device_map="cpu",
    torch_dtype=torch.float32,
    trust_remote_code=True,
    low_cpu_mem_usage=True,
)

print("Loading LoRA adapter...")
model = PeftModel.from_pretrained(model, LORA_PATH)

print("Merging LoRA weights...")
model = model.merge_and_unload()

tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)

model.eval()
print("Model & tokenizer loaded successfully")

# ─────────────────────────────────────────────
# Inference
# ─────────────────────────────────────────────
def generate_sql(question: str) -> str:
    if not question or not question.strip():
        return "Please enter a SQL-related question."

    messages = [
        {"role": "user", "content": question.strip()}
    ]

    input_ids = tokenizer.apply_chat_template(
        messages,
        tokenize=True,
        add_generation_prompt=True,
        return_tensors="pt",
    )

    with torch.inference_mode():
        output_ids = model.generate(
            input_ids=input_ids,
            max_new_tokens=MAX_NEW_TOKENS,
            temperature=TEMPERATURE,
            do_sample=DO_SAMPLE,
            pad_token_id=tokenizer.eos_token_id,
            use_cache=True,
        )

    response = tokenizer.decode(
        output_ids[0],
        skip_special_tokens=True
    )

    # Clean Phi-3 chat artifacts
    for token in ["<|assistant|>", "<|user|>", "<|end|>"]:
        if token in response:
            response = response.split(token)[-1]

    return response.strip() or "(empty response)"

# ─────────────────────────────────────────────
# Gradio UI
# ─────────────────────────────────────────────
demo = gr.Interface(
    fn=generate_sql,
    inputs=gr.Textbox(
        label="SQL Question",
        placeholder="Find duplicate emails in users table",
        lines=3,
    ),
    outputs=gr.Textbox(
        label="Generated SQL",
        lines=8,
    ),
    title="SQL Chat – Phi-3-mini (CPU)",
    description=(
        "CPU-only Hugging Face Space.\n"
        "First response may take 60–180 seconds. "
        "Subsequent requests are faster."
    ),
    examples=[
        ["Find duplicate emails in users table"],
        ["Top 5 highest paid employees"],
        ["Count orders per customer last month"],
        ["Delete duplicate rows based on email"],
    ],
    cache_examples=False,
)

# ─────────────────────────────────────────────
# Launch
# ─────────────────────────────────────────────
if __name__ == "__main__":
    print("Launching Gradio interface...")
    demo.launch(
        server_name="0.0.0.0",
        ssr_mode=False,   # important: avoids asyncio FD bug
        show_error=True,
    )