Update app.py

app.py

@@ -6,15 +6,18 @@ from sentence_transformers import SentenceTransformer, util
 from PIL import Image
 from typing import List
 import torch
-from transformers import BertTokenizer, BertModel
+from transformers import BertTokenizer, BertModel, T5ForConditionalGeneration, T5Tokenizer
 import torch.nn.functional as F
-import language_tool_python  # Import LanguageTool for grammar checking
 
 # Load pre-trained models
 tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
 bert_model = BertModel.from_pretrained('bert-base-uncased')
 sentence_model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
 
+# Load the pre-trained T5 model and tokenizer for grammar error detection
+grammar_model = T5ForConditionalGeneration.from_pretrained('t5-base')
+grammar_tokenizer = T5Tokenizer.from_pretrained('t5-base')
+
 # Initialize Groq client
 client = Groq(api_key=os.environ.get("GROQ_API_KEY"))
 
@@ -24,9 +27,6 @@ system_prompt = {
     "content": "You are a useful assistant. You reply with efficient answers."
 }
 
-# Initialize grammar checker
-tool = language_tool_python.LanguageTool('en-US')
-
 async def chat_groq(message, history):
     messages = [system_prompt]
     for msg in history:
@@ -103,13 +103,22 @@ def calculate_sentence_similarity(text1, text2):
     embedding2 = sentence_model.encode(text2, convert_to_tensor=True)
     return util.pytorch_cos_sim(embedding1, embedding2).item()
 
-def check_grammar(student_answer):
-    # Check grammar using LanguageTool
-    matches = tool.check(student_answer)
-    errors = len(matches)
+# Grammar detection and penalization using T5 model
+def detect_grammar_errors(text):
+    input_text = f"grammar: {text}"
+    inputs = grammar_tokenizer.encode(input_text, return_tensors='pt', max_length=512, truncation=True)
+    outputs = grammar_model.generate(inputs, max_length=512, num_beams=4, early_stopping=True)
+    grammar_analysis = grammar_tokenizer.decode(outputs[0], skip_special_tokens=True)
+    
+    # Count the number of errors based on specific indicators (customize based on analysis)
+    error_count = grammar_analysis.count('error')  # Use your own criteria
+    return error_count
+
+def penalize_for_grammar(student_answer):
+    grammar_errors = detect_grammar_errors(student_answer)
     
-    # Apply a penalty based on the number of grammar errors
-    penalty = 1 - min(0.1 * errors, 0.5)  # Maximum penalty is 50%
+    # Apply a penalty based on the number of grammar errors (max 50% penalty)
+    penalty = max(0.5, 1 - 0.05 * grammar_errors)
     return penalty
 
 def compare_answers(student_answer, teacher_answer):
@@ -120,7 +129,7 @@ def compare_answers(student_answer, teacher_answer):
     semantic_similarity = (0.75 * bert_similarity + 0.25 * sentence_similarity)
     
     # Apply grammar penalty
-    grammar_penalty = check_grammar(student_answer)
+    grammar_penalty = penalize_for_grammar(student_answer)
     final_similarity = semantic_similarity * grammar_penalty
     
     return final_similarity