update: readme about accelerate detail

app.py

@@ -146,7 +146,16 @@ with block:
     This allows generated images to maintain the same color composition as the original images. If you are looking to control both the contours and colors of the original image while using ControlNet to generate images, then this is the best option for you! You can try out this model or test the examples provided below 🤗.
 
     ## Update
-    Hi, everyone, We have added a Color-Canny-ControlNet accelerated version of our implementation based on Nvidia Triton and operator optimization. This faster ControlNet is deployed on a Nvidia A10 machine, and for a 512-pixel image, the inference takes about 1.2s, which is more faster than general implementation. 
+    Hi, everyone, We have added a Color-Canny-ControlNet accelerated version of our implementation based on Nvidia Triton and operator optimization. This faster ControlNet is deployed on a Nvidia A10 machine. For a 512-pixel image, the inference takes about 1.2s, which is more faster than general implementation with accelerate PyTorch2.0 about 40%.
+    We provide detailed test results as shown below.
+
+    | Method                | Infomation            | Inference Times       | Speed-up Ratio |
+    | --------------------- | --------------------- | --------------------- | -------------- |
+    | Benchmark             | [Huggingface implement](https://huggingface.co/blog/controlnet?spm=ata.21736010.0.0.422d24288Kj7zm) | 3.00s(V100)/5.00s(T4) | /              |
+    | Accelerate PyTorch2.0 | xFormers              | 2.03s(A10)            | 0              |
+    |                       | SDPA                  | 2.02s(A10)            | 0.5%           |
+    | Ours                  | TRT & OP optimize     | 1.20s(A10)            | 40.4%          |
+
     Welcome to try this [faster Color-Canny-ControlNet](http://121.40.118.209:7860/).
     """)
     with gr.Row():