app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import os

# === Model ID and Token ===
model_id = "TrabbyPatty/mistral-7b-instruct-finetuned-flashcards-4bit"
hf_token = os.getenv("alluse")  # Hugging Face token from Space secrets

# === Load tokenizer & model with authentication ===
tokenizer = AutoTokenizer.from_pretrained(
    model_id,
    token=hf_token,
    use_fast=False   # ✅ force slow tokenizer (fixes JSON error)
)

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",       # let HF map to GPU/CPU
    torch_dtype=torch.float16,
    token=hf_token
)

# === SYSTEM MESSAGE ===
SYSTEM_MESSAGE = """<<SYS>>
You are a strict flashcard generator.
- Only extract information from the input.
- Do NOT add outside knowledge, assumptions, or details not mentioned in the input.
- Always follow the requested format exactly.
<</SYS>>"""

def generate_flashcards(user_input, max_new_tokens=600, temperature=0.5):
    # Format the prompt with system + user input
    prompt = (
        f"<s>[INST] {SYSTEM_MESSAGE}\n\n"
        f"Create flashcards strictly using only the information provided.\n\n"
        f"Input: {user_input}[/INST]\nOutput:"
    )

    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=max_new_tokens,
            temperature=temperature,
            do_sample=False,
            repetition_penalty=1.05,
            pad_token_id=tokenizer.eos_token_id,
            eos_token_id=tokenizer.eos_token_id,
        )

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

    # Extract only the Output section
    if "Output:" in response:
        final_answer = response.split("Output:")[-1].strip()
    else:
        final_answer = response.strip()

    return final_answer


# ✅ Gradio UI
demo = gr.Interface(
    fn=generate_flashcards,
    inputs=[
        gr.Textbox(label="Enter study text", lines=8, placeholder="Paste your study material here..."),
        gr.Slider(100, 1000, value=600, step=50, label="Max New Tokens"),
        gr.Slider(0.1, 1.0, value=0.5, step=0.1, label="Temperature"),
    ],
    outputs="text",
    title="Flashcard Generator (Mistral-7B LoRA)",
    description="Paste study material and generate flashcards. Model strictly extracts only from input."
)

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