Update app.py

app.py

@@ -3,7 +3,7 @@ import torch
 import numpy as np
 from transformers import AutoTokenizer, AutoModelForSequenceClassification
 
-# --- Configuration ---
+# Configuration 
 # Your Model's Hugging Face Repository ID
 # NOTE: This assumes you successfully pushed the model during the CV loop!
 HF_REPO_ID = "sumittguptaa148/DL-Gen-AI-Project"
@@ -20,19 +20,12 @@ BEST_THRESHOLDS = {
     'surprise': 0.48
 }
 
-# --- Load Model and Tokenizer ---
+# Load Model and Tokenizer 
 try:
-    # We load AutoModel for a custom architecture, but since we pushed
-    # using Trainer, the hub might save it with an AutoModelForSequenceClassification structure
-    # Let's try to load the base AutoModel and then load the state dict manually
-    
     # Load the tokenizer
     tokenizer = AutoTokenizer.from_pretrained(HF_REPO_ID)
 
     # Load the model weights and structure.
-    # We use AutoModelForSequenceClassification here as Hugging Face Trainer usually
-    # saves a classification head that is compatible with this class structure
-    # when pushing to the hub.
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     model = AutoModelForSequenceClassification.from_pretrained(HF_REPO_ID, num_labels=len(LABELS))
     model.to(device)
@@ -47,7 +40,7 @@ except Exception as e:
     device = "cpu"
 
 
-# --- Prediction Function ---
+# Prediction Function 
 def predict_emotion(text):
     if model is None:
         return "Model failed to load. Please check logs."
@@ -78,32 +71,25 @@ def predict_emotion(text):
         threshold = BEST_THRESHOLDS[label]
         
         # Classification
-        is_present = " Present" if prob >= threshold else " Absent"
+        # Using unicode checkmark/cross for better visual appeal in the Gradio app
+        is_present = "✅ Present" if prob >= threshold else "❌ Absent" 
         
         # Format for display
         results[f"{label.capitalize()} ({is_present})"] = f"{prob:.4f} (Threshold: {threshold:.2f})"
         
     return results
 
-# --- Gradio Interface ---
-# Output components for Gradio
-output_components = [
-    gr.Textbox(label=f"{label.capitalize()} (Classification & Prob)", lines=1) for label in LABELS
-]
-
-# Map output components to keys in the dictionary returned by predict_emotion
-# Gradio expects the output component labels to match the dictionary keys
-output_keys = [f"{label.capitalize()} ({{}})" for label in LABELS] # Placeholder for "Present/Absent" part
+# Gradio Interface Setup 
 
 # Custom function to create the correct output components for Gradio
 def get_output_components():
     # Use Textbox for results
     outputs = []
     for label in LABELS:
+        # NOTE: Changing label slightly to avoid conflict with the key returned by predict_emotion()
         outputs.append(gr.Textbox(label=f"{label.capitalize()} Emotion Result", lines=1))
     return outputs
 
-
 # Custom wrapper to ensure output matches the order of Gradio components
 def predict_emotion_gradio(text):
     raw_results = predict_emotion(text)
@@ -117,7 +103,6 @@ def predict_emotion_gradio(text):
         
     return tuple(ordered_results)
 
-
 title = "Multi-Label Emotion Classification with RoBERTa-Large"
 description = "A DL/GenAI project classifying text into Anger, Fear, Joy, Sadness, and Surprise. The model uses a fine-tuned RoBERTa-Large with 5-Fold CV and dynamic threshold optimization."
 
@@ -127,6 +112,5 @@ gr.Interface(
     outputs=get_output_components(), # Use the custom function to get ordered components
     title=title,
     description=description,
-    # allow_flagging="never",
-    theme="huggingface"
+    # The 'allow_flagging' and 'theme' arguments have been removed to fix the TypeErrors.
 ).launch()