Upload folder using huggingface_hub

app.py

@@ -1,8 +1,10 @@
 import gradio as gr
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
 import torch
+import numpy as np
 
 MODEL_NAME = "changcheng967/Aegis-Qwen3-1.7B-SFT"
+DETECTOR_NAME = "desklib/ai-text-detector-v1.01"
 
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained(
@@ -12,6 +14,14 @@ model = AutoModelForCausalLM.from_pretrained(
     trust_remote_code=True,
 )
 
+detector = pipeline(
+    "text-classification",
+    model=DETECTOR_NAME,
+    device_map="auto",
+    truncation=True,
+    max_length=512,
+)
+
 SYSTEM_PROMPT = (
     "You are a student rewriting text in your own natural voice. "
     "Rewrite the following AI-generated text to sound like a real student wrote it. "
@@ -20,9 +30,38 @@ SYSTEM_PROMPT = (
 )
 
 
+def compute_stats(text):
+    sentences = [s.strip() for s in text.replace("!", ".").replace("?", ".").split(".") if s.strip()]
+    if not sentences:
+        return {"burstiness": 0, "avg_sentence_len": 0, "sentence_len_std": 0, "vocab_diversity": 0}
+
+    lengths = [len(s.split()) for s in sentences]
+    words = text.lower().split()
+    unique = len(set(words))
+
+    return {
+        "burstiness": float(np.std(lengths) / max(np.mean(lengths), 1)),
+        "avg_sentence_len": round(np.mean(lengths), 1),
+        "sentence_len_std": round(np.std(lengths), 1),
+        "vocab_diversity": round(unique / max(len(words), 1), 3),
+    }
+
+
+def score_text(text):
+    try:
+        result = detector(text)[0]
+        if result["label"] == "HUMAN":
+            human_score = 1 - result["score"]
+        else:
+            human_score = result["score"]
+        return round(human_score * 100, 1)
+    except Exception:
+        return -1
+
+
 def humanize(ai_text, temperature, max_tokens):
     if not ai_text.strip():
-        return "Please enter some AI-generated text."
+        return "", "", ""
 
     messages = [
         {"role": "system", "content": SYSTEM_PROMPT},
@@ -42,7 +81,24 @@ def humanize(ai_text, temperature, max_tokens):
         )
 
     response = tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True)
-    return response
+
+    input_score = score_text(ai_text)
+    output_score = score_text(response)
+
+    input_stats = compute_stats(ai_text)
+    output_stats = compute_stats(response)
+
+    report = f"--- Detection Scores ---\n"
+    report += f"Input (AI):   {input_score}% likely human\n"
+    report += f"Output:       {output_score}% likely human\n"
+    report += f"\n--- Statistical Features ---\n"
+    report += f"{'Metric':<22} {'Input':>8} {'Output':>8}\n"
+    report += f"{'Burstiness':<22} {input_stats['burstiness']:>8.2f} {output_stats['burstiness']:>8.2f}\n"
+    report += f"{'Avg sentence length':<22} {input_stats['avg_sentence_len']:>8.1f} {output_stats['avg_sentence_len']:>8.1f}\n"
+    report += f"{'Sentence length std':<22} {input_stats['sentence_len_std']:>8.1f} {output_stats['sentence_len_std']:>8.1f}\n"
+    report += f"{'Vocab diversity':<22} {input_stats['vocab_diversity']:>8.3f} {output_stats['vocab_diversity']:>8.3f}\n"
+
+    return response, report
 
 
 demo = gr.Interface(
@@ -52,9 +108,12 @@ demo = gr.Interface(
         gr.Slider(0.1, 1.5, value=0.7, step=0.1, label="Temperature"),
         gr.Slider(128, 1024, value=512, step=64, label="Max Tokens"),
     ],
-    outputs=gr.Textbox(label="Humanized Text", lines=8),
+    outputs=[
+        gr.Textbox(label="Humanized Text", lines=8),
+        gr.Textbox(label="Detection Report", lines=10),
+    ],
     title="Aegis Humanizer",
-    description="Rewrites AI-generated text to sound like a real student wrote it. Powered by Aegis Qwen3-1.7B SFT.",
+    description="Rewrites AI-generated text to sound like a real student. Shows detection scores and statistical features.",
     examples=[
         ["Artificial intelligence has significantly impacted the field of education. It provides personalized learning experiences and helps teachers identify areas where students need improvement.", 0.7, 512],
         ["Climate change is one of the most pressing issues facing the world today. Rising temperatures, melting ice caps, and extreme weather events are all consequences of human activity.", 0.7, 512],

requirements.txt

@@ -1,3 +1,4 @@
 transformers>=4.40.0
 torch>=2.0.0
 accelerate>=0.20.0
+numpy