Merge pull request #58 from BBuf/patch-1

Avoid computing model initialization time

tests/test_inference.py

@@ -137,6 +137,7 @@ def main():
     if args.quantize:
         model = quantize(model, weight_bit_width=8, backend="torch")
     model.cuda()
+    torch.cuda.synchronize()
     
     with open(args.prompt_file, "r") as f:
         prompt = f.readlines()