app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import os

# --- 1. Securely Get the Hugging Face Token from Space Secrets ---
auth_token = os.getenv("HF_TOKEN")

# --- DEBUGGING STEP: Check if the token was found ---
# This will print to your Space's logs.
if auth_token is not None:
    # Do not print the full token for security. Just confirm it was found.
    print("✅ HF_TOKEN secret was found by the application.")
else:
    print("❌ HF_TOKEN secret was NOT found. Please double-check it is set in your Space Settings.")
# --- END DEBUGGING STEP ---


# --- 2. Load your Model using the Token ---
MODEL_ID = "Bur3hani/karani_ofline"
MODEL_LOADED = False

try:
    if auth_token:
        print(f"Attempting to load tokenizer and model from Hub: {MODEL_ID}...")
        tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, token=auth_token)
        model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_ID, token=auth_token)
        print("✅ Model and Tokenizer loaded successfully.")
        MODEL_LOADED = True
    else:
        # This will be printed if the secret is missing.
        print("Skipping model loading because HF_TOKEN is missing.")

except Exception as e:
    print(f"❌ An error occurred while loading the model from the Hub: {e}")


# --- 3. Define the Prediction Function ---
def get_chat_response(message, history):
    if not MODEL_LOADED:
        return "ERROR: The AI model failed to load. Please check the Space logs for the exact error."

    input_text = ""
    for user_turn, bot_turn in history:
        input_text += f"user: {user_turn} bot: {bot_turn} "
    input_text += f"user: {message}"

    inputs = tokenizer(input_text, return_tensors="pt", max_length=512, truncation=True)
    outputs = model.generate(**inputs, max_length=60, num_beams=4, early_stopping=True)
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response

# --- 4. Build and Launch the Gradio Interface ---
demo = gr.ChatInterface(
    fn=get_chat_response,
    title="Karani v1 - AI Secretary",
    description="A conversational AI assistant for Kiswahili, powered by a custom fine-tuned model.",
    examples=[["Habari za asubuhi?"], ["Ni nini mpango wa leo?"]],
)

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