app.py

import gradio as gr
from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
import json

model_name = "google/flan-t5-small"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
generator = pipeline("text2text-generation", model=model, tokenizer=tokenizer)

def generate_json(prompt):
    # Make the instruction explicit: return ONLY JSON, without explanation
    instruction = (
        f"Generate only a valid JSON object without any markdown or additional text. "
        f"The JSON object should have the following keys: title, author, and tags. "
        f"Fill in dummy values. Now create a JSON object for: {prompt}"
    )
    result = generator(instruction, max_length=256, do_sample=False)
    generated_text = result[0]["generated_text"].strip()  # remove extra whitespace
    
    print(f"Raw Model Output: {generated_text}")  # Debug statement
    
    try:
        parsed = json.loads(generated_text)
        formatted_json = json.dumps(parsed, indent=2)
    except Exception as e:
        formatted_json = f"Raw Output:\n{generated_text}\n\nError parsing JSON: {e}"
    
    return formatted_json

demo = gr.Interface(
    fn=generate_json,
    inputs=gr.Textbox(lines=4, label="Enter Prompt"),
    outputs=gr.Textbox(lines=20, label="Generated JSON"),
    title="Lightweight JSON Generator",
    description="Enter a prompt describing the structure or content you want in JSON format."
)

demo.queue()
demo.launch(show_error=True)