Update app.py

app.py

@@ -7,6 +7,8 @@ from models.tokenize import Tokenizer
 from models.llama import *
 from generate import LmGeneration
 
+import os
+os.environ['CUDA_LAUNCH_BLOCKING'] = '1'
 
 args = None
 lm_generation = None
@@ -22,7 +24,7 @@ def init_args():
     args.batch_size = 1
     args.seq_length = 512
     args.world_size = 1
-    args.use_int8 = True
+    args.use_int8 = False
     args.top_p = 0
     args.repetition_penalty_range = 1024
     args.repetition_penalty_slope = 0