app3.py

import gradio as gr
from transformers import GPT2LMHeadModel, GPT2Tokenizer, TextDataset, DataCollatorForLanguageModeling, Trainer, TrainingArguments
from datasets import load_dataset
import numpy as np
import torch

# Load GPT2 Model and Tokenizer
model_name = "gpt2"
tokenizer = GPT2Tokenizer.from_pretrained(model_name)
model = GPT2LMHeadModel.from_pretrained(model_name)

# Define PPO Training Function (simplified)
def fine_tune_gpt2_with_ppo(dataset_name, epochs, learning_rate):
    # Load the dataset
    dataset = load_dataset(dataset_name)
    
    # Prepare dataset for GPT-2 training
    def encode(examples):
        return tokenizer(examples['text'], truncation=True, padding='max_length', max_length=128)

    tokenized_dataset = dataset.map(encode, batched=True)
    train_dataset = tokenized_dataset["train"]
    
    # Prepare data collator and training arguments
    data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False)
    training_args = TrainingArguments(
        output_dir="./results",
        overwrite_output_dir=True,
        num_train_epochs=epochs,
        per_device_train_batch_size=4,
        save_steps=10_000,
        save_total_limit=2,
        learning_rate=learning_rate
    )
    
    # Trainer
    trainer = Trainer(
        model=model,
        args=training_args,
        data_collator=data_collator,
        train_dataset=train_dataset
    )
    
    # Train model
    trainer.train()
    
    return "Training Completed!"

# Gradio Interface
def train_interface(dataset, epochs, learning_rate):
    result = fine_tune_gpt2_with_ppo(dataset, int(epochs), float(learning_rate))
    return result

# Gradio App
gradio_interface = gr.Interface(
    fn=train_interface,
    inputs=[
        gr.inputs.Textbox(label="Dataset (e.g. 'wikitext')"),
        gr.inputs.Slider(1, 10, step=1, label="Epochs"),
        gr.inputs.Textbox(label="Learning Rate")
    ],
    outputs="text",
    title="GPT-2 RL Training App",
    description="Fine-tune GPT-2 using PPO via a Gradio interface."
)

# Launch the app
gradio_interface.launch()