app.py

import gradio as gr

def generate_text(context, num_samples, context_length, model_name):
    from base import main
    from pathlib import Path
    
    if model_name == "pythia_160m_deduped_custom" or model_name == "pythia_160m_deduped_huggingface":
        ckpt_dir = Path('checkpoints/EleutherAI/pythia-160m-deduped')
    elif model_name == "pythia_70m_deduped":
        ckpt_dir = Path('checkpoints/EleutherAI/pythia-70m-deduped')
    elif model_name == "pythia_410m_deduped":
        ckpt_dir = Path('checkpoints/EleutherAI/pythia-410m-deduped')        
        
    context = str(context)
    num_samples = int(num_samples)
    context_length = int(context_length)
    model_name = str(model_name)

    output_msg_list = main(prompt=context, checkpoint_dir=ckpt_dir, model_name=model_name, num_samples=num_samples, max_new_tokens=context_length)
    output_msg = str()
    for idx, msg in enumerate(output_msg_list):
        title = f"--Generated message : {idx + 1} using the model : {model_name}--\n"
        output_msg += f"{title}\n"
        output_msg += f"{msg}\n"
        output_msg += f"\n"
    return output_msg

def gradio_fn(context, num_samples, context_length, model_name):
    output_txt_msg = generate_text(context, num_samples, context_length, model_name)    
    return output_txt_msg

markdown_description = """
- This is a Gradio app that generates text based on
  - given text context
  - for given character length
  - number of Samples
  - using Selected GPT model
- Currently following models are available :
  - **(a)** pythia_160m_deduped_huggingface **(b)** pythia_160m_deduped_custom \
    **(c)** pythia_410m_deduped **(d)** pythia_70m_deduped
"""
demo = gr.Interface(fn=gradio_fn,
                    inputs=[gr.Textbox(info="Start my passage with: 'I would like to'"),
                            gr.Number(value=1, minimum=1, maximum=5, \
                                      info="Number of samples to be generated min=1, max=5"),
                            gr.Slider(value=50, minimum=50, maximum=250, \
                                      info="Num characters for passage min=50, max=250"),  
                            gr.Dropdown(["pythia_160m_deduped_huggingface", "pythia_160m_deduped_custom",
                                         "pythia_410m_deduped", "pythia_70m_deduped"], \
                                        multiselect=False, label="Model-Name", \
                                        info="Pretrained model to be used for text generation")],                       
                    outputs=gr.Textbox(),
                    title="DialogGen - Dialogue Generator",
                    description=markdown_description,
                    article=" **Credits** : https://github.com/Lightning-AI/lit-gpt ")

demo.launch(share=True)