update perf (#313)

Author: hjchen2

README.md

@@ -12,7 +12,7 @@ Updated on Nov 6, 2023.
 > | 3090           | 42.38it/s       | 42.33it/s       | 6.66it/s                  |
 > | 4090           | 74.71it/s       | 73.57it/s       | 13.57it/s                 |
 > | A100-PCIE-40GB | 54.4it/s        | 54.06it/s       | 10.22it/s                 |
-> | A100-SXM4-80GB | 58.68it/s       | 58.02it/s       | 11.80it/s                 |
+> | A100-SXM4-80GB | 59.68it/s       | 61.91it/s       | 11.80it/s                 |
 
 
 ## Business inquiry

examples/text_to_image.py

@@ -1,6 +1,6 @@
 import argparse
 from onediff.infer_compiler import oneflow_compile
-from onediff import EulerDiscreteScheduler
+from onediff import EulerDiscreteScheduler, rewrite_self_attention
 from diffusers import StableDiffusionPipeline
 import oneflow as flow
 import torch
@@ -35,8 +35,9 @@ def parse_args():
     torch_dtype=torch.float16,
     safety_checker=None,
 )
-
 pipe = pipe.to("cuda")
+
+rewrite_self_attention(pipe.unet)
 pipe.unet = oneflow_compile(pipe.unet)
 
 prompt = args.prompt