import os import gradio as gr from transformers import MT5Tokenizer, MT5ForConditionalGeneration # Load your fine-tuned mT5 model model_name = "Addaci/mT5-small-experiment-13-checkpoint-2790" tokenizer = MT5Tokenizer.from_pretrained(model_name) model = MT5ForConditionalGeneration.from_pretrained(model_name) def correct_htr(raw_htr_text): try: inputs = tokenizer(raw_htr_text, return_tensors="pt", max_length=512, truncation=True) outputs = model.generate(**inputs, max_length=128, num_beams=4, early_stopping=True) corrected_text = tokenizer.decode(outputs[0], skip_special_tokens=True) return corrected_text except Exception as e: return str(e) def summarize_text(legal_text): try: inputs = tokenizer("summarize: " + legal_text, return_tensors="pt", max_length=512, truncation=True) outputs = model.generate(**inputs, max_length=150, num_beams=4, early_stopping=True) summary = tokenizer.decode(outputs[0], skip_special_tokens=True) return summary except Exception as e: return str(e) def answer_question(legal_text, question): try: formatted_input = f"question: {question} context: {legal_text}" inputs = tokenizer(formatted_input, return_tensors="pt", max_length=512, truncation=True) outputs = model.generate(**inputs, max_length=150, num_beams=4, early_stopping=True) answer = tokenizer.decode(outputs[0], skip_special_tokens=True) return answer except Exception as e: return str(e) # Create the Gradio Blocks interface with boxed clickable buttons and bold text with gr.Blocks() as demo: gr.Markdown("# mT5 Legal Assistant") gr.Markdown("Use this tool to correct raw HTR, summarize legal texts, or answer questions about legal cases.") # Adding external link buttons with a box around them and bold text with gr.Row(): gr.HTML('''
''') with gr.Tab("Correct HTR"): gr.Markdown("### Correct Raw HTR Text") raw_htr_input = gr.Textbox(lines=5, placeholder="Enter raw HTR text here...") corrected_output = gr.Textbox(lines=5, placeholder="Corrected HTR text") correct_button = gr.Button("Correct HTR") clear_button = gr.Button("Clear") correct_button.click(correct_htr, inputs=raw_htr_input, outputs=corrected_output) clear_button.click(lambda: ("", ""), outputs=[raw_htr_input, corrected_output]) with gr.Tab("Summarize Legal Text"): gr.Markdown("### Summarize Legal Text") legal_text_input = gr.Textbox(lines=10, placeholder="Enter legal text to summarize...") summary_output = gr.Textbox(lines=5, placeholder="Summary of legal text") summarize_button = gr.Button("Summarize Text") clear_button = gr.Button("Clear") summarize_button.click(summarize_text, inputs=legal_text_input, outputs=summary_output) clear_button.click(lambda: ("", ""), outputs=[legal_text_input, summary_output]) with gr.Tab("Answer Legal Question"): gr.Markdown("### Answer a Question Based on Legal Text") legal_text_input_q = gr.Textbox(lines=10, placeholder="Enter legal text...") question_input = gr.Textbox(lines=2, placeholder="Enter your question...") answer_output = gr.Textbox(lines=5, placeholder="Answer to your question") answer_button = gr.Button("Get Answer") clear_button = gr.Button("Clear") answer_button.click(answer_question, inputs=[legal_text_input_q, question_input], outputs=answer_output) clear_button.click(lambda: ("", "", ""), outputs=[legal_text_input_q, question_input, answer_output]) # Launch the Gradio interface demo.launch()