app.py

import gradio as gr
import json
from transformers import AutoModelForSequenceClassification, Trainer, TrainingArguments, AutoTokenizer
from datasets import Dataset
import shutil
import os

# Load tokenized data
def load_data(tokenized_file):
    with open(tokenized_file.name, 'r') as f:
        tokenized_data = json.load(f)
    return tokenized_data

# Fine-tune the model
def fine_tune_model(tokenized_file, progress=gr.Progress()):
    tokenized_data = load_data(tokenized_file)
    
    # Convert tokenized data to Dataset
    dataset = Dataset.from_dict(tokenized_data)
    
    # Split the dataset into train and validation sets
    tokenized_datasets = dataset.train_test_split(test_size=0.2)
    
    model = AutoModelForSequenceClassification.from_pretrained('anferico/bert-for-patents', num_labels=2)
    tokenizer = AutoTokenizer.from_pretrained('anferico/bert-for-patents')
    
    training_args = TrainingArguments(
        output_dir='./results',
        num_train_epochs=3,
        per_device_train_batch_size=8,  # Reduce batch size
        per_device_eval_batch_size=16,
        gradient_accumulation_steps=2,  # Use gradient accumulation
        fp16=True,  # Enable mixed precision training
        warmup_steps=500,
        weight_decay=0.01,
        logging_dir='./logs',
        logging_steps=10,
        evaluation_strategy="epoch",
        save_strategy="epoch",
        load_best_model_at_end=True,
    )
    
    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=tokenized_datasets['train'],
        eval_dataset=tokenized_datasets['test'],
    )
    
    progress(0.5, "Fine-tuning the model...")
    trainer.train()
    progress(1.0, "Fine-tuning complete.")
    
    model.save_pretrained('./fine_tuned_patentbert')
    tokenizer.save_pretrained('./fine_tuned_patentbert')
    
    # Create a zip file of the fine-tuned model
    shutil.make_archive('fine_tuned_patentbert', 'zip', './fine_tuned_patentbert')
    
    return "Model fine-tuned and saved successfully. Download the model using the link below.", "fine_tuned_patentbert.zip"

# Create Gradio interface
iface = gr.Interface(
    fn=fine_tune_model,
    inputs=[
        gr.File(label="Upload Tokenized Data JSON")
    ],
    outputs=[
        gr.Textbox(label="Processing Information"),
        gr.File(label="Download Fine-Tuned Model")
    ],
    title="Fine-Tune Patent BERT Model",
    description="Upload tokenized JSON file to fine-tune the BERT model.",
    live=True  # Enable live updates for progress
)

# Launch the interface
iface.launch()