Update app.py

app.py

@@ -1,14 +1,6 @@
-from transformers import T5ForConditionalGeneration, T5Tokenizer, AutoModel, AutoTokenizer
-import torch
-from sklearn.metrics.pairwise import cosine_similarity
-import numpy as np
-import gradio as gr
-from collections import Counter
-import pandas as pd
-
 # Load paraphrase model and tokenizer
 model = T5ForConditionalGeneration.from_pretrained('ramsrigouthamg/t5_paraphraser')
-tokenizer = T5Tokenizer.from_pretrained('t5-base', legacy=False)  # Explicitly set legacy=False
+tokenizer = T5Tokenizer.from_pretrained('t5-base')
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 model = model.to(device)
@@ -27,9 +19,6 @@ def get_sentence_embedding(sentence):
 
 # Paraphrasing function
 def paraphrase_sentence(sentence):
-    if not sentence.strip():
-        return "Please enter a valid sentence."
-    
     # Updated prompt for statement-like output
     text = "rephrase as a statement: " + sentence
     encoding = tokenizer.encode_plus(text, padding=False, return_tensors="pt")
@@ -40,10 +29,10 @@ def paraphrase_sentence(sentence):
         attention_mask=attention_masks,
         do_sample=True,
         max_length=128,          
-        top_k=40,                
-        top_p=0.85,              
+        top_k=40,                 # Reduced top_k for less randomness
+        top_p=0.85,               # Reduced top_p for focused sampling
         early_stopping=True,
-        num_return_sequences=5   
+        num_return_sequences=5    # Generate 5 paraphrases
     )
 
     # Decode and format paraphrases with numbering
@@ -69,16 +58,14 @@ def calculate_precision_recall_accuracy(sentences):
         original_tokens = Counter(sentence.lower().split())
         
         for paraphrase in paraphrases:
-            if not paraphrase.strip():
-                continue
             # Remove numbering before evaluation
-            paraphrase_text = paraphrase.split(". ", 1)[1] if ". " in paraphrase else paraphrase
-            paraphrase_embedding = get_sentence_embedding(paraphrase_text)
+            paraphrase = paraphrase.split(". ", 1)[1]
+            paraphrase_embedding = get_sentence_embedding(paraphrase)
             similarity = cosine_similarity(original_embedding.cpu(), paraphrase_embedding.cpu())[0][0]
             total_similarity += similarity
 
             # Calculate precision and recall based on token overlap
-            paraphrase_tokens = Counter(paraphrase_text.lower().split())
+            paraphrase_tokens = Counter(paraphrase.lower().split())
             overlap = sum((paraphrase_tokens & original_tokens).values())
             precision = overlap / sum(paraphrase_tokens.values()) if paraphrase_tokens else 0
             recall = overlap / sum(original_tokens.values()) if original_tokens else 0
@@ -88,159 +75,34 @@ def calculate_precision_recall_accuracy(sentences):
             paraphrase_count += 1
 
     # Calculate averages for accuracy, precision, and recall
-    overall_accuracy = (total_similarity / paraphrase_count) * 100 if paraphrase_count else 0
-    avg_precision = (total_precision / paraphrase_count) * 100 if paraphrase_count else 0
-    avg_recall = (total_recall / paraphrase_count) * 100 if paraphrase_count else 0
-
-    return (f"**Overall Model Accuracy (Semantic Similarity):** {overall_accuracy:.2f}%\n"
-            f"**Average Precision (Token Overlap):** {avg_precision:.2f}%\n"
-            f"**Average Recall (Token Overlap):** {avg_recall:.2f}%")
-
-# Custom CSS for aesthetic design
-custom_css = """
-body {
-    background: linear-gradient(135deg, #e0e7ff, #c3dafe, #e0e7ff);
-    font-family: 'Inter', sans-serif;
-}
-.gradio-container {
-    max-width: 800px !important;
-    margin: auto;
-    padding: 20px;
-    background: white;
-    border-radius: 20px;
-    box-shadow: 0 10px 30px rgba(0, 0, 0, 0.1);
-}
-h1 {
-    font-size: 2.5rem;
-    font-weight: 700;
-    background: linear-gradient(to right, #4f46e5, #7c3aed);
-    -webkit-background-clip: text;
-    -webkit-text-fill-color: transparent;
-    text-align: center;
-    margin-bottom: 1rem;
-}
-textarea, input {
-    border: 2px solid #e0e7ff !important;
-    border-radius: 10px !important;
-    padding: 15px !important;
-    transition: all 0.3s ease !important;
-}
-textarea:hover, input:hover {
-    border-color: #a5b4fc !important;
-    box-shadow: 0 0 10px rgba(79, 70, 229, 0.2) !important;
-}
-textarea:focus, input:focus {
-    border-color: #4f46e5 !important;
-    box-shadow: 0 0 15px rgba(79, 70, 229, 0.3) !important;
-}
-button {
-    background: linear-gradient(to right, #4f46e5, #7c3aed) !important;
-    color: white !important;
-    font-weight: 600 !important;
-    padding: 12px 24px !important;
-    border-radius: 10px !important;
-    border: none !important;
-    transition: all 0.3s ease !important;
-}
-button:hover {
-    background: linear-gradient(to right, #4338ca, #6d28d9) !important;
-    transform: scale(1.05) !important;
-    box-shadow: 0 5px 15px rgba(79, 70, 229, 0.4) !important;
-}
-button:disabled {
-    background: linear-gradient(to right, #a3a3a3, #d1d5db) !important;
-    transform: none !important;
-    box-shadow: none !important;
-}
-.output-text {
-    background: #f9fafb !important;
-    border-radius: 10px !important;
-    padding: 15px !important;
-    border: 1px solid #e5e7eb !important;
-    transition: all 0.3s ease !important;
-}
-.output-text:hover {
-    background: #eff6ff !important;
-    border-color: #a5b4fc !important;
-}
-footer {
-    display: none !important;
-}
-"""
-
-# Custom JavaScript for additional interactivity
-custom_js = """
-<script>
-document.addEventListener('DOMContentLoaded', () => {
-    const textarea = document.querySelector('textarea');
-    const button = document.querySelector('button');
-    
-    // Add typing animation effect
-    textarea.addEventListener('input', () => {
-        textarea.style.transform = 'scale(1.02)';
-        setTimeout(() => {
-            textarea.style.transform = 'scale(1)';
-        }, 200);
-    });
-    
-    // Button click animation
-    button.addEventListener('click', () => {
-        if (!button.disabled) {
-            button.style.transform = 'scale(0.95)';
-            setTimeout(() => {
-                button.style.transform = 'scale(1)';
-            }, 200);
-        }
-    });
-});
-</script>
-"""
-
-# Define Gradio UI with enhanced aesthetics
-with gr.Blocks(theme=gr.themes.Soft(), css=custom_css, js=custom_js) as demo:
-    gr.Markdown(
-        """
-        # PARA-GEN: Aesthetic Paraphraser
-        Enter a sentence below to generate five beautifully rephrased statements.
-        """
-    )
-    
-    with gr.Row():
-        with gr.Column(scale=3):
-            input_text = gr.Textbox(
-                label="Input Sentence",
-                placeholder="Type your sentence here...",
-                lines=4,
-                max_lines=4
-            )
-            paraphrase_button = gr.Button("Generate Paraphrases")
-        
-        with gr.Column(scale=2):
-            output_text = gr.Textbox(
-                label="Paraphrased Results",
-                lines=10,
-                interactive=False
-            )
-    
-    with gr.Accordion("Model Performance Metrics", open=False):
-        metrics_output = gr.Markdown()
-    
-    # Define button click behavior
-    paraphrase_button.click(
-        fn=paraphrase_sentence,
-        inputs=input_text,
-        outputs=output_text
-    )
-    
-    # Calculate and display metrics on load without _js
-    test_sentences = [
-        "The quick brown fox jumps over the lazy dog.",
-        "Artificial intelligence is transforming industries.",
-        "The weather is sunny and warm today.",
-        "He enjoys reading books on machine learning.",
-        "The stock market fluctuates daily due to various factors."
-    ]
-    metrics_output.value = calculate_precision_recall_accuracy(test_sentences)
-
-# Launch Gradio app
-demo.launch(share=False)
+    overall_accuracy = (total_similarity / paraphrase_count) * 100
+    avg_precision = (total_precision / paraphrase_count) * 100
+    avg_recall = (total_recall / paraphrase_count) * 100
+
+    print(f"Overall Model Accuracy (Semantic Similarity): {overall_accuracy:.2f}%")
+    print(f"Average Precision (Token Overlap): {avg_precision:.2f}%")
+    print(f"Average Recall (Token Overlap): {avg_recall:.2f}%")
+
+# Define Gradio UI
+iface = gr.Interface(
+    fn=paraphrase_sentence,
+    inputs="text",
+    outputs="text",
+    title="PARA-GEN (T5 Paraphraser)",
+    description="Enter a sentence, and the model will generate five numbered paraphrases in statement form."
+)
+
+# List of test sentences to evaluate metrics
+test_sentences = [
+    "The quick brown fox jumps over the lazy dog.",
+    "Artificial intelligence is transforming industries.",
+    "The weather is sunny and warm today.",
+    "He enjoys reading books on machine learning.",
+    "The stock market fluctuates daily due to various factors."
+]
+
+# Calculate overall accuracy, precision, and recall for the list of test sentences
+calculate_precision_recall_accuracy(test_sentences)
+
+# Launch Gradio app (Gradio UI will not show metrics)
+iface.launch(share=False)