changes made to app.py

app.py

@@ -96,16 +96,24 @@ def identify_subtopic(question, domain):
 
 def generate_question(prompt, domain, state=None):
     full_prompt = system_prompt + "\n" + prompt
-    inputs = tokenizer(full_prompt, return_tensors="pt").to(device)
+    # Explicitly set padding side and add pad token
+    tokenizer.padding_side = "left"
+    if tokenizer.pad_token is None:
+        tokenizer.pad_token = tokenizer.eos_token
+    
+    # Tokenize with explicit padding and attention mask
+    inputs = tokenizer(full_prompt, return_tensors="pt", padding=True, truncation=True).to(device)
+    
     outputs = model.generate(
         inputs["input_ids"],
+        attention_mask=inputs["attention_mask"],
         max_new_tokens=50,
         no_repeat_ngram_size=2,
         top_k=30,
         top_p=0.9,
         temperature=0.7,
         do_sample=True,
-        pad_token_id=tokenizer.eos_token_id,
+        pad_token_id=tokenizer.pad_token_id,
     )
     question = tokenizer.decode(outputs[0], skip_special_tokens=True).strip()
     if not question.endswith("?"):
@@ -124,6 +132,11 @@ def generate_question(prompt, domain, state=None):
 
 
 def evaluate_response(response, question):
+    # Explicitly set padding side and add pad token
+    qwq_tokenizer.padding_side = "left"
+    if qwq_tokenizer.pad_token is None:
+        qwq_tokenizer.pad_token = qwq_tokenizer.eos_token
+    
     eval_prompt = (
         "Evaluate the following candidate response to an interview question.\n\n"
         f"**Question:** {question}\n"
@@ -132,15 +145,19 @@ def evaluate_response(response, question):
         "Also, provide a brief suggestion for improvement. Format:\n"
         "Rating: <Rating>\nSuggestion: <Suggestion>"
     )
-    inputs = qwq_tokenizer(eval_prompt, return_tensors="pt", padding=True).to(qwq_model.device)
+    
+    # Tokenize with explicit padding and attention mask
+    inputs = qwq_tokenizer(eval_prompt, return_tensors="pt", padding=True, truncation=True).to(qwq_model.device)
+    
     outputs = qwq_model.generate(
         inputs["input_ids"],
+        attention_mask=inputs["attention_mask"],
         max_new_tokens=100,
         top_k=30,
         top_p=0.9,
         temperature=0.7,
         do_sample=True,
-        pad_token_id=qwq_tokenizer.eos_token_id,
+        pad_token_id=qwq_tokenizer.pad_token_id,
     )
     evaluation = qwq_tokenizer.decode(outputs[0], skip_special_tokens=True)
     rating, suggestion = "Unknown", "No suggestion available."