Update app.py

app.py

@@ -12,7 +12,7 @@ import torch.nn.functional as F
 # 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')
+sentence_model = SentenceTransformer('paraphrase-MiniLM-L6-v2')
 
 # Load the pre-trained T5 model and tokenizer for grammar error detection
 grammar_model = T5ForConditionalGeneration.from_pretrained('t5-base')
@@ -118,7 +118,7 @@ def penalize_for_grammar(student_answer):
     grammar_errors = detect_grammar_errors(student_answer)
     
     # Apply a penalty based on the number of grammar errors (max 50% penalty)
-    penalty = max(0.5, 1 - 0.05 * grammar_errors)
+    penalty = max(0.3, 1 - 0.1 * grammar_errors)
     return penalty
 
 def compare_answers(student_answer, teacher_answer):
@@ -126,7 +126,7 @@ def compare_answers(student_answer, teacher_answer):
     sentence_similarity = calculate_sentence_similarity(student_answer, teacher_answer)
     
     # Use a higher weight for BERT similarity
-    semantic_similarity = (0.75 * bert_similarity + 0.25 * sentence_similarity)
+    semantic_similarity = (0.50 * bert_similarity + 0.50 * sentence_similarity)
     
     # Apply grammar penalty
     grammar_penalty = penalize_for_grammar(student_answer)