Update app.py

app.py

@@ -1,7 +1,7 @@
+import gradio as gr
 from flask import Flask, request
 from transformers import RobertaForSequenceClassification, RobertaTokenizer, RobertaConfig
 import torch
-import gradio as gr
 import os
 import re
 app = Flask(__name__)
@@ -13,6 +13,11 @@ model = RobertaForSequenceClassification.from_pretrained("PirateXX/ChatGPT_Detec
 model_name = "roberta-base"
 tokenizer = RobertaTokenizer.from_pretrained(model_name, map_location=torch.device('cpu'))
 
+device = 'cuda' if cuda.is_available() else 'cpu'
+model_id = "gpt2"
+modelgpt2 = GPT2LMHeadModel.from_pretrained(model_id).to(device)
+tokenizergpt2 = GPT2TokenizerFast.from_pretrained(model_id)
+
 def text_to_sentences(text):
     clean_text = text.replace('\n', ' ')
     return re.split(r'(?<=[^A-Z].[.?]) +(?=[A-Z])', clean_text)
@@ -64,8 +69,54 @@ def findRealProb(text):
     realProb = ans/cnt
     return {"Real": realProb, "Fake": 1-realProb}, results
 
+
+def text_to_sentences(text):
+    clean_text = text.replace('\n', ' ')
+    return re.split(r'(?<=[^A-Z].[.?]) +(?=[A-Z])', clean_text)
+
+def calculatePerplexity(text):
+    encodings = tokenizergpt2("\n\n".join([text]), return_tensors="pt")
+    max_length = modelgpt2.config.n_positions
+    stride = 512
+    seq_len = encodings.input_ids.size(1)
+
+    nlls = []
+    prev_end_loc = 0
+    for begin_loc in range(0, seq_len, stride):
+        end_loc = min(begin_loc + max_length, seq_len)
+        trg_len = end_loc - prev_end_loc
+        input_ids = encodings.input_ids[:, begin_loc:end_loc].to(device)
+        target_ids = input_ids.clone()
+        target_ids[:, :-trg_len] = -100
+
+        with torch.no_grad():
+            outputs = modelgpt2(input_ids, labels=target_ids)
+            neg_log_likelihood = outputs.loss * trg_len
+
+        nlls.append(neg_log_likelihood)
+
+        prev_end_loc = end_loc
+        if end_loc == seq_len:
+            break
+
+    ppl = torch.exp(torch.stack(nlls).sum() / end_loc)
+
+    return ppl.item()
+    
+@app.get("/getPerplexities")
+def calculatePerplexities(text):
+    sentences = text_to_sentences(text)
+    perplexities = []
+    for sentence in sentences:
+        perplexity = calculatePerplexity(sentence)
+        label = "Human"
+        if perplexity<25:
+            label = "AI"
+        perplexities.append({"sentence": sentence, "perplexity": perplexity, "label": label})
+    return perplexities
+
 demo = gr.Interface(
-        fn=[findRealProb, predict], 
+        fn=findRealProb, 
         inputs=gr.Textbox(placeholder="Copy and paste here..."), 
          article = "Visit <a href = \"https://ai-content-detector.online/\">AI Content Detector</a> for better user experience!",
         outputs=gr.outputs.JSON(),