import spaces import gradio as gr from wtpsplit import SaT import json # Initialize the SaT model sat = SaT("sat-12l-sm") sat.half().to("cuda") @spaces.GPU(duration=59) def segment_text(input_text, multi_doc_input): results = {} if input_text: # Process single text input sentences = sat.split(input_text) results["input_text"] = {"segments": sentences} if multi_doc_input: try: # Parse the JSON input documents = json.loads(multi_doc_input) for key, doc in documents.items(): sentences = sat.split(doc) results[f"doc_{key}"] = {"segments": sentences} except json.JSONDecodeError: results["error"] = "Invalid JSON input for multiple documents" # Create a JSON object with the results json_output = json.dumps(results, indent=2) return json_output # Create the Gradio interface iface = gr.Interface( fn=segment_text, inputs=[ gr.Textbox(lines=5, label="Input Text (Optional)"), gr.Textbox(lines=10, label="Multiple Documents JSON (Optional)", placeholder='{"1": "Document 1 text", "2": "Document 2 text"}') ], outputs=gr.JSON(label="Segmented Text (JSON)"), title="Text Segmentation with SaT", description="This app uses the SaT (Segment any Text) model to split input text into sentences and return the result as JSON. You can input text directly or provide multiple documents as JSON. All credits to the respective author(s). Github: https://github.com/segment-any-text/wtpsplit/tree/main", examples=[ ["This is a test This is another test.", ""], ["Hello this is a test But this is different now Now the next one starts looool", ""], ["The quick brown fox jumps over the lazy dog It was the best of times, it was the worst of times", ""], ["", '{"1": "Document 1 first sentence Document 1 second sentence", "2": "Document 2 only sentence", "3": "Document 3 first Document 3 second"}'] ] ) # Launch the app iface.launch()