Update app.py

app.py

@@ -1,13 +1,55 @@
 import gradio as gr
-from transformers import pipeline
+import torch
+from transformers import AutoTokenizer, AutoModelForTokenClassification
 
-ner = pipeline(
-    "token-classification",
-    model="techysanoj/fine-tuned-IndicNER",
-    aggregation_strategy="simple"
-)
+MODEL_ID = "techysanoj/fine-tuned-IndicNER"
 
-def predict(text):
-    return ner(text)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+model = AutoModelForTokenClassification.from_pretrained(MODEL_ID)
 
-gr.Interface(fn=predict, inputs="text", outputs="json").launch()
+id2label = {int(k): v for k, v in model.config.id2label.items()}
+
+
+def ner_predict(text):
+    # tokenize input
+    inputs = tokenizer(text, return_tensors="pt")
+    tokens = tokenizer.convert_ids_to_tokens(inputs["input_ids"][0])
+
+    # run model
+    with torch.no_grad():
+        logits = model(**inputs).logits
+
+    pred_ids = torch.argmax(logits, dim=-1)[0].tolist()
+
+    # build output table
+    rows = []
+    for tok, pid in zip(tokens, pred_ids):
+        rows.append([tok, id2label[pid]])
+
+    # pretty text version
+    pretty_output = ""
+    for tok, lab in rows:
+        pretty_output += f"{tok:15} → {lab}\n"
+
+    return pretty_output, rows
+
+
+# gradio UI
+with gr.Blocks(title="Indic NER Token-wise Output") as demo:
+    gr.Markdown("🔥 Indian Language NER — Token Level Output (Hindi + English)")
+    
+    inp = gr.Textbox(lines=3, label="Enter text")
+    
+    btn = gr.Button("Run NER")
+    
+    out_text = gr.Textbox(label="Tokenized Output")
+    out_table = gr.Dataframe(
+        headers=["Token", "Label"], 
+        datatype=["str", "str"], 
+        label="Table View",
+        wrap=True
+    )
+
+    btn.click(fn=ner_predict, inputs=inp, outputs=[out_text, out_table])
+
+demo.launch()