add caching for preds

app.py

@@ -148,7 +148,7 @@ def highlight_text(text, ner_threshold, re_threshold):
             "end": end
         })
 
-    return {"text": text, "entities": entities}
+    return {"text": text, "entities": entities}, {"ner": ner_preds, "relations": rel_preds}
 
 # JSON output function
 
@@ -166,9 +166,12 @@ def get_model_predictions(text, ner_threshold, re_threshold):
     ner_preds, rel_preds = prune_acronym_and_self_relations(ner_preds, rel_preds)
     return json.dumps({"ner": ner_preds, "relations": rel_preds}, indent=2)
 
-# Build Gradio UI
-demo = gr.Blocks()
-with demo:
+def _cached_predictions(state):
+    if not state:
+        return "📋 No predictions yet. Click **Submit** first."
+    return json.dumps(state, indent=2)
+
+with gr.Blocks as demo:
     gr.Markdown("## Data Use Detector\n"
                 "Adjust the sliders below to set thresholds, then:\n"
                 "- **Submit** to highlight entities.\n"
@@ -195,18 +198,18 @@ with demo:
     txt_out = gr.HighlightedText(label="Annotated Entities")
 
     get_pred_btn = gr.Button("Get Model Predictions")
-    ner_rel_box = gr.Textbox(label="Model Predictions (JSON)", lines=15)
-
+    json_out      = gr.Textbox(label="Model Predictions (JSON)", lines=15)
+    state         = gr.State()
     # Wire up interactions
     highlight_btn.click(
         fn=highlight_text,
         inputs=[txt_in, ner_slider, re_slider],
-        outputs=txt_out
+        outputs=[txt_out, state]
     )
     get_pred_btn.click(
-        fn=get_model_predictions,
-        inputs=[txt_in, ner_slider, re_slider],
-        outputs=ner_rel_box
+        fn=_cached_predictions,
+        inputs=[state],
+        outputs=[json_out]
     )
 
     # Enable queue for concurrency