app.py

import os
import gradio as gr
from transformers import pipeline

hf_token = os.getenv("hf_token")
# Initialize the text generation pipeline
generator = pipeline("text-generation", model="isitcoding/gpt2_120_finetuned", use_auth_token=hf_token)

# Define the response function with additional options for customization
def text_generation(
    prompt: str,
    details: bool = False,
    stream: bool = False,
    model: str = None,
    best_of: int = None,
    decoder_input_details: bool = None,
    do_sample: bool = False,
    frequency_penalty: float = None,
    grammar: None = None,
    max_new_tokens: int = None,
    repetition_penalty: float = None
):
    # Setup the configuration for the model generation
    gen_params = {
        "max_length": 518,  # Default, you can tweak it or set from parameters
        "num_return_sequences": 1,
        "do_sample": do_sample,
        "temperature": 0.7,  # Controls randomness
        "top_k": 50,  # You can adjust for more control over sampling
        "top_p": 0.9,  # Same as above, for sampling
    }

    if max_new_tokens:
        gen_params["max_length"] = max_new_tokens + len(prompt.split())

    if frequency_penalty:
        gen_params["frequency_penalty"] = frequency_penalty

    if repetition_penalty:
        gen_params["repetition_penalty"] = repetition_penalty

    # Generate the text based on the input prompt and parameters
    generated_text = generator(prompt, **gen_params)[0]["generated_text"]

    if details:
        # Return additional details for debugging if needed
        return {
            "generated_text": generated_text,
            "params_used": gen_params
        }
    else:
        return generated_text

# Create Gradio interface
iface = gr.Interface(
    fn=text_generation,  # The function we defined
    inputs=[
        gr.Textbox(label="Input Prompt"),  # User input prompt
        gr.Checkbox(label="Show Details", default=False),  # Option for additional details
        gr.Checkbox(label="Stream Mode", default=False),  # Streaming checkbox (not used in this example)
        gr.Textbox(label="Model (optional)", default=None),  # Optional model name
        gr.Slider(minimum=1, maximum=5, label="Best of (Optional)", default=None),
        gr.Slider(minimum=0.0, maximum=2.0, label="Frequency Penalty (Optional)", default=None),
        gr.Slider(minimum=0.0, maximum=2.0, label="Repetition Penalty (Optional)", default=None),
    ],
    outputs="text"  # Output is plain text
)

# Launch the interface
iface.launch()