app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import random
from datasets import load_dataset
from transformers import GPT2LMHeadModel, GPT2Tokenizer

# Load dataset
dataset = load_dataset("rahul7star/hindi-poetry")["train"]

# Load your model and tokenizer
model_name = "rahul7star/hindi_poetry_language_model"
tokenizer = GPT2Tokenizer.from_pretrained(model_name)
model = GPT2LMHeadModel.from_pretrained(model_name)

# 2️⃣ Function to Generate Hindi Poetry
def generate_poetry_base(prompt, max_length=100, temperature=0.7, top_k=50, top_p=0.95):
    input_ids = tokenizer(prompt, return_tensors="pt").input_ids
    with torch.no_grad():
        output = model.generate(
            input_ids, 
            max_length=max_length, 
            temperature=temperature, 
            top_k=top_k, 
            top_p=top_p,
            pad_token_id=tokenizer.pad_token_id
        )
    return tokenizer.decode(output[0], skip_special_tokens=True)

def generate_poetry(prompt, max_length=100, temperature=0.7, top_k=50, top_p=0.95):
    input_ids = tokenizer(prompt, return_tensors="pt").input_ids
    with torch.no_grad():
        output = model.generate(
            input_ids,
            max_length=max_length,
            temperature=temperature,  # Increased randomness
            top_p=top_p,
            do_sample=True,
            repetition_penalty=1.5,  # Added repetition penalty to prevent duplicates
            num_beams=5,  # Use beam search for higher quality output
            no_repeat_ngram_size=2,  # Prevent repeating the same n-grams
            early_stopping=True,
            pad_token_id=tokenizer.pad_token_id
        )
    return tokenizer.decode(output[0], skip_special_tokens=True)

# Poetry Generation Function with Random Selection from Dataset and Explicit 4-Line Structure
def generate_random_poem(prompt, max_length=180, temperature=1.0, top_p=0.9, num_lines=4):
    # Randomly select a line from the dataset
    random_line = random.choice(dataset["poem"])

    # Prepare the input text with the random line selected, and start with a unique phrase to avoid repetition
    input_text = f"{random_line} "  # Unique start to force variety

    # Tokenize the input text
    encoding = tokenizer(input_text, return_tensors="pt", padding=True, truncation=True, max_length=max_length)
    input_ids = encoding.input_ids.to(model.device)
    attention_mask = encoding.attention_mask.to(model.device)

    # Set pad_token_id to eos_token_id
    pad_token_id = tokenizer.eos_token_id

    # Generate the poem using the model with increased randomness
    output = model.generate(
        input_ids,
        attention_mask=attention_mask,
        max_length=max_length,
        temperature=temperature,  # Increased randomness
        top_p=top_p,
        do_sample=True,
        repetition_penalty=1.5,  # Added repetition penalty to prevent duplicates
        num_beams=5,  # Use beam search for higher quality output
        no_repeat_ngram_size=2,  # Prevent repeating the same n-grams
        early_stopping=True,
        pad_token_id=pad_token_id
    )

    # Decode the output and split into lines
    generated_poem = tokenizer.decode(output[0], skip_special_tokens=True)
    generated_poem = generated_poem.strip()

    # Split the generated text into separate lines based on full stops (Hindi poems often end with "।")
    poem_lines = generated_poem.split("।")

   

    final_poem = "\n".join(poem_lines)

    return final_poem

# 3️⃣ Gradio Interface
interface = gr.Interface(
    fn=generate_random_poem,  # Use function directly without calling it
    inputs=[
        gr.Textbox(label="Enter Prompt", placeholder="Start your Hindi poem..."),
        gr.Slider(50, 500, step=10, value=180, label="Max Length"),
        gr.Slider(0.1, 1.5, step=0.1, value=0.7, label="Temperature"),
        gr.Slider(1, 100, step=1, value=50, label="Top-k Sampling"),
        gr.Slider(0.1, 1.0, step=0.05, value=0.95, label="Top-p Sampling"),
    ],
    outputs=gr.Textbox(label="Generated Hindi Poem"),
    title="Hindi Poetry Generator ✨",
    description="Generate beautiful Hindi poetry. Just enter a prompt and adjust parameters. Example: 'मैया मोरी'",
)

# 4️⃣ Run the Gradio App
interface.launch(share=True)