Update app.py

app.py

@@ -2,28 +2,36 @@ import gradio as gr
 import spaces
 from transformers import pipeline
 
-# Load pipeline
-model_id = "HaniaEmaan/code-switching-codesaviours-si26-hania"
+model_id = "your-actual-username/code-switching-codesaviours-si26-hania"
+
+# Set aggregation_strategy to None to stop merging words into single chunks
 nlp_pipeline = pipeline(
     "token-classification", 
     model=model_id, 
-    aggregation_strategy="simple", 
+    aggregation_strategy=None, 
     ignore_labels=[]
 )
 
-# The @spaces.GPU decorator dynamically requests GPU compute for inference
 @spaces.GPU
 def classify_text(text):
-    predictions = nlp_pipeline(text)
-    formatted_output = []
-    for p in predictions:
-        formatted_output.append(f"Word/Phrase: '{p['word']}' | Label: {p['entity_group']} | Score: {p['score']:.2f}")
-    return "\n".join(formatted_output)
+    # Split input by words to classify each token individually
+    words = text.split()
+    results = []
+    
+    for word in words:
+        predictions = nlp_pipeline(word)
+        if predictions:
+            # Pick the top token prediction for each individual word
+            top_pred = predictions[0]
+            label = top_pred.get('entity', top_pred.get('entity_group', 'UNK'))
+            score = top_pred['score']
+            results.append(f"Word: '{word}' | Label: {label} | Score: {score:.2f}")
+            
+    return "\n".join(results)
 
-# Gradio Interface
 demo = gr.Interface(
     fn=classify_text,
-    inputs=gr.Textbox(placeholder="Enter a sentence... e.g., Photo postain checko"),
+    inputs=gr.Textbox(placeholder="Enter text..."),
     outputs="textbox",
     title="Code-Switched Language Identification (ZeroGPU)",
     description="XLM-RoBERTa Token Classification powered by Hugging Face ZeroGPU."