Update app.py

app.py

@@ -87,9 +87,10 @@ RAW NOTE:
         with torch.no_grad():
             outputs = model.generate(
                 **inputs, 
-                max_new_tokens=256, 
+                max_new_tokens=64, 
                 temperature=0.1,
-                do_sample=True
+                do_sample=True,
+                use_cache=False
             )
         
         text = tokenizer.batch_decode(outputs)[0]
@@ -143,10 +144,11 @@ PRIVATE_DB:
         with torch.no_grad():
             outputs = model.generate(
                 **inputs, 
-                max_new_tokens=1024, 
+                max_new_tokens=256, 
                 temperature=0.1, 
                 repetition_penalty=1.05,
-                do_sample=True
+                do_sample=True,
+                use_cache=False
             )
         
         text = tokenizer.batch_decode(outputs)[0]