Update app.py

app.py

@@ -8,8 +8,8 @@ import gradio as gr
 # -----------------------------
 # MODEL INITIALIZATION
 # -----------------------------
-# This is a DeBERTa-v3-Large model fine-tuned on the DAIGT (Student Writing vs AI) dataset.
-MODEL_NAME = "Hamidreza/DeBERTa-v3-large-AI-Detector-v2"
+# desklib/ai-text-detector-v1.01 is highly robust for academic/essay detection.
+MODEL_NAME = "desklib/ai-text-detector-v1.01"
 tokenizer = None
 model = None
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
@@ -18,7 +18,9 @@ def get_model():
     global tokenizer, model
     if model is None:
         print(f"Loading High-Performance Model: {MODEL_NAME} on {device}")
-        # DeBERTa-v3 requires use_fast=False for stable SentencePiece tokenization
+        
+        # DeBERTa-v3 requires use_fast=False for stable SentencePiece tokenization.
+        # Ensure 'sentencepiece' is installed (pip install sentencepiece).
         tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, use_fast=False)
         
         dtype = torch.float32
@@ -93,7 +95,7 @@ def analyze(text):
     if not pure_sents:
         return "—", "—", "<em>No sentences detected.</em>", None
 
-    # Sliding window inference (Contextual)
+    # Sliding window inference (Contextual for better accuracy)
     windows = []
     for i in range(len(pure_sents)):
         start = max(0, i - 1)
@@ -102,6 +104,7 @@ def analyze(text):
 
     inputs = tok(windows, return_tensors="pt", padding=True, truncation=True, max_length=512).to(device)
     logits = mod(**inputs).logits
+    # Note: Desklib uses Label 1 for AI-generated and Label 0 for Human.
     probs = F.softmax(logits.float(), dim=-1)[:, 1].cpu().numpy().tolist()
 
     lengths = [len(s.split()) for s in pure_sents]
@@ -124,9 +127,9 @@ def analyze(text):
             
             # Binary logic: Threshold applied to color
             if score >= THRESHOLD:
-                color, bg = "#b80d0d", "rgba(184, 13, 13, 0.15)" # RED
+                color, bg = "#b80d0d", "rgba(184, 13, 13, 0.15)" # RED (AI)
             else:
-                color, bg = "#11823b", "rgba(17, 130, 59, 0.15)" # GREEN
+                color, bg = "#11823b", "rgba(17, 130, 59, 0.15)" # GREEN (Human)
             
             highlighted_html += (
                 f"<span style='background:{bg}; padding:2px 4px; border-radius:4px; border-bottom: 2px solid {color};' "
@@ -152,8 +155,8 @@ def analyze(text):
 # GRADIO INTERFACE
 # -----------------------------
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.Markdown("## 🕵️ AI Detector Pro (DeBERTa Edition)")
-    gr.Markdown(f"Advanced Academic Analysis. Threshold: **{THRESHOLD*100:.0f}%**. Everything below is categorized as Human.")
+    gr.Markdown("## 🕵️ AI Detector Pro (Academic Edition)")
+    gr.Markdown(f"Using **{MODEL_NAME}**. Threshold: **{THRESHOLD*100:.0f}%**. Scores below this are marked as Human.")
     
     with gr.Row():
         with gr.Column(scale=3):