app.py

import os
import gradio as gr
import torch
from transformers import AutoTokenizer, MistralForCausalLM
from huggingface_hub import snapshot_download
from mcp_serpapi import search_pdf_url
from utils import download_pdf

# ---------------------------
# Config
# ---------------------------
HF_TOKEN = os.environ.get("HF_TOKEN")  # stored in Spaces secrets

MODEL_REPO = "mistralai/Mistral-7B-Instruct-v0.3"
MODEL_PATH = "/tmp/mistral"
CACHE_DIR = "/tmp/.cache/huggingface"
OFFLOAD_DIR = "/tmp/offload"

os.makedirs(MODEL_PATH, exist_ok=True)
os.makedirs(CACHE_DIR, exist_ok=True)
os.makedirs(OFFLOAD_DIR, exist_ok=True)
os.environ["HF_HOME"] = CACHE_DIR

# ---------------------------
# Download model if needed
# ---------------------------
if not os.listdir(MODEL_PATH):
    print("Downloading Mistral model snapshot to /tmp...")
    snapshot_download(
        repo_id=MODEL_REPO,
        repo_type="model",
        local_dir=MODEL_PATH,
        use_auth_token=HF_TOKEN
    )
else:
    print("Model already exists in /tmp. Skipping download.")

# ---------------------------
# Load tokenizer and Mistral model
# ---------------------------
print("Loading tokenizer and MistralForCausalLM model...")

tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, use_fast=False)

model = MistralForCausalLM.from_pretrained(
    MODEL_PATH,
    device_map="auto",
    offload_folder=OFFLOAD_DIR,
    torch_dtype=torch.float16
)

# ---------------------------
# Model query function
# ---------------------------
def query_mistral(prompt: str) -> str:
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    outputs = model.generate(
        **inputs,
        max_new_tokens=512,
        do_sample=True,
        temperature=0.7
    )
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# ---------------------------
# PDF search function
# ---------------------------
def find_and_download_annual_report(company: str) -> str:
    query = f"site:example.com {company} annual report filetype:pdf"
    pdf_url = search_pdf_url(query)
    if pdf_url:
        local_path = download_pdf(pdf_url, save_dir="/tmp/pdfs")
        return f"Downloaded: {local_path}"
    return "No PDF found"

# ---------------------------
# Gradio UI
# ---------------------------
with gr.Blocks() as demo:
    gr.Markdown("## Mistral-7B-Instruct Chatbot + Annual Report PDF Finder")

    with gr.Tab("Chat with Mistral"):
        prompt_box = gr.Textbox(label="Enter prompt")
        mistral_output = gr.Textbox(label="Model output")
        mistral_button = gr.Button("Run")

        mistral_button.click(
            fn=query_mistral,
            inputs=prompt_box,
            outputs=mistral_output
        )

    with gr.Tab("Download Annual Report PDF"):
        company_box = gr.Textbox(label="Company Name")
        pdf_output = gr.Textbox(label="Result")
        pdf_button = gr.Button("Find PDF")

        pdf_button.click(
            fn=find_and_download_annual_report,
            inputs=company_box,
            outputs=pdf_output
        )

demo.launch(server_name="0.0.0.0", server_port=7860)