app.py

import torch
from fastai.text.all import *
from huggingface_hub import hf_hub_download

# Step 1: Recreate the DataLoaders used during training
def create_dataloaders():
    data = [
        {"input": "Who is rahul7star?", "output": "Rahul7star is a software developer living in NSW."},
        {"input": "What is Rahul7star's profession?", "output": "Rahul7star is an AI creator and software developer."},
        {"input": "What does Rahul7star do?", "output": "Rahul7star works as a developer and enjoys solving complex coding problems."},
        {"input": "Tell me about Rahul7star", "output": "Rahul7star is a talented developer, AI creator, and a fan of traveling."},
        {"input": "What is Rahul7star known for?", "output": "Rahul7star is known for his work in AI, software development, and his love for coding."}
    ]
    
    # Split into training and validation sets
    train_data = data[:4]  # First 4 examples for training
    valid_data = data[4:]  # Last example for validation

    # Define the DataBlock (use the same structure as during training)
    dblock = DataBlock(
        blocks=(TextBlock.from_df(text_cols='input'), CategoryBlock),
        get_x=lambda x: x['input'],
        get_y=lambda x: x['output'],
        splitter=RandomSplitter(valid_pct=0.2, seed=42)
    )

    # Create DataLoaders
    dls = dblock.dataloaders(train_data, bs=2)  # Batch size of 2 for quick experimentation
    return dls

# Step 2: Load the model weights
def load_model_weights():
    model_path = hf_hub_download(
        repo_id="rahul7star/fastai-rahul-text-model-v02", 
        filename="rahul9star.pth"
    )
    
    # Load the model architecture (you need to define the same architecture used during training)
    learn = text_classifier_learner(create_dataloaders(), AWD_LSTM, metrics=accuracy)
    
    # Load the weights into the model
    learn.load(model_path)
    print("Model loaded successfully from Hugging Face.")
    return learn

# Step 3: Use the model in your Gradio interface
learn = load_model_weights()

# Check if the model is loaded successfully
if learn is None:
    raise ValueError("Failed to load the model")

# Step 4: Define the Gradio Interface
def predict(input_text):
    try:
        # Get prediction from the model
        pred, _, probs = learn.predict(input_text)
        return f"Prediction: {pred}, Confidence: {probs.max():.2f}"
    except Exception as e:
        return f"Error during prediction: {e}"

# Step 5: Create Gradio Interface
import gradio as gr
gr.Interface(fn=predict, inputs="text", outputs="text").launch()