Update app.py

app.py

@@ -69,30 +69,30 @@ def process_audio(microphone, audio_upload, state, answer_mode):  # Added audio_
         You answer questions clearly and simply, using age-appropriate language.
         You are also a little bit silly and like to make jokes."""
     prompt = f"{system_prompt}\nUser: {text}"
-    #with torch.no_grad():
-    vicuna_input = vicuna_tokenizer(prompt, return_tensors="pt").to('cuda')
-    if answer_mode == 'slow':
-        vicuna_output = vicuna_model.generate(
-            **vicuna_input,
-            max_length = 512,
-            min_new_tokens = 256,
-            do_sample = True
-        )
-    if answer_mode == 'medium':
-        vicuna_output = vicuna_model.generate(
-            **vicuna_input,
-            max_length = 128,
-            min_new_tokens = 64,
-            do_sample = True
-        )
-    if answer_mode == 'fast':
-        vicuna_output = vicuna_model.generate(
-            **vicuna_input,
-            max_length = 42,
-            min_new_tokens = 16,
-            do_sample = True
-        )
-    vicuna_response = vicuna_tokenizer.decode(vicuna_output[0], skip_special_tokens=True)
+    with torch.no_grad():
+        vicuna_input = vicuna_tokenizer(prompt, return_tensors="pt").to('cuda')
+        if answer_mode == 'slow':
+            vicuna_output = vicuna_model.generate(
+                **vicuna_input,
+                max_length = 512,
+                min_new_tokens = 256,
+                do_sample = True
+            )
+        if answer_mode == 'medium':
+            vicuna_output = vicuna_model.generate(
+                **vicuna_input,
+                max_length = 128,
+                min_new_tokens = 64,
+                do_sample = True
+            )
+        if answer_mode == 'fast':
+            vicuna_output = vicuna_model.generate(
+                **vicuna_input,
+                max_length = 42,
+                min_new_tokens = 16,
+                do_sample = True
+            )
+        vicuna_response = vicuna_tokenizer.decode(vicuna_output[0], skip_special_tokens=True)
     vicuna_response = vicuna_response.replace(prompt, "").strip()
     updated_state = state + "\nUser: " + text + "\n" + "Tutor: " + vicuna_response
     try: