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


# Load pre-trained GPT-2 model and tokenizer
model_name = "gpt2-large"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

def generate_text(input_text, max_length=16, num_beams=5, do_sample=False, no_repeat_ngram_size=2):
    """
    Generate text based on the given input text.
    Parameters:
    - input_text (str): The input text to start generation from.
    - max_length (int): Maximum length of the generated text.
    - num_beams (int): Number of beams for beam search.
    - do_sample (bool): Whether to use sampling or not.
    - no_repeat_ngram_size (int): Size of the n-gram to avoid repetition.
    Returns:
    - generated_text (str): The generated text.
    """
    # Encode the input text and move it to the appropriate device
    input_ids = tokenizer(input_text, return_tensors='pt')['input_ids']
    # Generate text using the model
    output = model.generate(input_ids, max_length=max_length, num_beams=num_beams,
                            do_sample=do_sample, no_repeat_ngram_size=no_repeat_ngram_size)
    # Decode the generated output
    generated_text = tokenizer.decode(output[0])
    return generated_text

# def generate_text_with_nucleus_search(input_text, max_length=16, do_sample=True, top_p=0.9):
#     """
#     Generate text with nucleus sampling based on the given input text.
#     Parameters:
#     - input_text (str): The input text to start generation from.
#     - max_length (int): Maximum length of the generated text.
#     - do_sample (bool): Whether to use sampling or not.
#     - top_p (float): Nucleus sampling parameter.
#     Returns:
#     - generated_text (str): The generated text.
#     """
#     # Encode the input text and move it to the appropriate device
#     input_ids = tokenizer(input_text, return_tensors='pt')['input_ids']
#     # Generate text using nucleus sampling
#     output = model.generate(input_ids, max_length=max_length, do_sample=do_sample, top_p=top_p)
#     # Decode the generated output
#     generated_text = tokenizer.decode(output[0])
#     return generated_text

# Create Gradio input interface
input_text_interface = gr.Textbox(lines=5, label="Input Text", placeholder="Enter text for generation...")

# Create Gradio output interface for regular text generation
output_text_interface1 = gr.Textbox(label="Generated Text (Regular)", placeholder="Generated text will appear here...")

# Interface for regular text generation
interface1 = gr.Interface(generate_text, input_text_interface, output_text_interface1, 
                          title="Text Generation with GPT-2",
                          description="Generate text using the GPT-2 model with regular generation method.",
                          allow_flagging="never")

# Create Gradio output interface for text generation with nucleus sampling
# output_text_interface2 = gr.Textbox(label="Generated Text (Nucleus Sampling)", placeholder="Generated text will appear here...")

# # Interface for text generation with nucleus sampling
# interface2 = gr.Interface(generate_text_with_nucleus_search, input_text_interface, output_text_interface2, 
#                           title="Text Generation with Nucleus Sampling",
#                           description="Generate text using nucleus sampling with the GPT-2 model.",
#                           allow_flagging="never")

# Launch both interfaces
interface1.launch(share=True)
# interface2.launch(share=True)