Create app.py

app.py

@@ -0,0 +1,71 @@
+import gradio as gr
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+from sentence_transformers import SentenceTransformer, util
+import torch
+import torch.nn.functional as F
+
+# Load SBERT model
+sbert_model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
+
+# Load NLI model
+nli_model_name = "tasksource/ModernBERT-base-nli"
+nli_tokenizer = AutoTokenizer.from_pretrained(nli_model_name)
+nli_model = AutoModelForSequenceClassification.from_pretrained(nli_model_name)
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+nli_model.to(device)
+
+# SBERT function
+def compute_similarity(text1, text2):
+    embeddings = sbert_model.encode([text1, text2], convert_to_tensor=True)
+    similarity = float(util.pytorch_cos_sim(embeddings[0], embeddings[1])[0])
+    
+    interpretation = ""
+    if similarity > 0.9:
+        interpretation = "🟢 Very High Similarity"
+    elif similarity > 0.75:
+        interpretation = "🟡 Moderate Similarity"
+    elif similarity > 0.5:
+        interpretation = "🟠 Low Similarity"
+    else:
+        interpretation = "🔴 Very Low Similarity"
+    
+    return round(similarity, 4), interpretation
+
+# NLI function
+def check_entail(premise, hypothesis):
+    inputs = nli_tokenizer(premise, hypothesis, return_tensors="pt", truncation=True, padding=True, max_length=512).to(device)
+    with torch.no_grad():
+        logits = nli_model(**inputs).logits
+    probs = torch.softmax(logits, dim=-1)[0]
+    label = ["entailment", "neutral", "contradiction"][torch.argmax(probs).item()]
+    return label, { "entailment": float(probs[0]), "neutral": float(probs[1]), "contradiction": float(probs[2]) }
+
+def run_bi_direction(a, b):
+    res1 = check_entail(a, b)
+    res2 = check_entail(b, a)
+    return res1[0], res1[1], res2[0], res2[1]
+
+# Build the interface
+with gr.Blocks() as demo:
+    gr.Markdown("# ✍️ Essay Comparison Tool")
+
+    with gr.Tab("Semantic Similarity (SBERT)"):
+        a1 = gr.Textbox(label="Essay A", lines=8)
+        b1 = gr.Textbox(label="Essay B", lines=8)
+        sim_button = gr.Button("Compare Similarity")
+        sim_score = gr.Textbox(label="Cosine Similarity (0–1)")
+        sim_interpret = gr.Textbox(label="Interpretation")
+        sim_button.click(fn=compute_similarity, inputs=[a1, b1], outputs=[sim_score, sim_interpret])
+
+    with gr.Tab("Bidirectional Entailment (NLI)"):
+        a2 = gr.Textbox(label="Essay A (Original)", lines=8)
+        b2 = gr.Textbox(label="Essay B (Modified)", lines=8)
+        nli_button = gr.Button("Run Entailment Check")
+        ab_label = gr.Textbox(label="A ➜ B Label")
+        ab_scores = gr.JSON(label="A ➜ B Scores")
+        ba_label = gr.Textbox(label="B ➜ A Label")
+        ba_scores = gr.JSON(label="B ➜ A Scores")
+        nli_button.click(fn=run_bi_direction, inputs=[a2, b2], outputs=[ab_label, ab_scores, ba_label, ba_scores])
+
+if __name__ == "__main__":
+    demo.launch()