update doc
Browse files
README.md
CHANGED
@@ -14,11 +14,11 @@ tags:
|
|
14 |
- text-to-image
|
15 |
---
|
16 |
|
17 |
-
# Stable Diffusion XL Turbo for ONNX Runtime
|
18 |
|
19 |
## Introduction
|
20 |
|
21 |
-
This repository hosts the optimized
|
22 |
|
23 |
The models are generated by [Olive](https://github.com/microsoft/Olive/tree/main/examples/stable_diffusion) with command like the following:
|
24 |
```
|
@@ -34,18 +34,18 @@ See the [usage instructions](#usage-example) for how to run the SDXL pipeline wi
|
|
34 |
- **License:** [STABILITY AI NON-COMMERCIAL RESEARCH COMMUNITY LICENSE](https://huggingface.co/stabilityai/sd-turbo/blob/main/LICENSE)
|
35 |
- **Model Description:** This is a conversion of the [SD-Turbo](https://huggingface.co/stabilityai/sd-turbo) model for [ONNX Runtime](https://github.com/microsoft/onnxruntime) inference with CUDA execution provider.
|
36 |
|
37 |
-
## Performance
|
38 |
|
39 |
#### Latency
|
40 |
|
41 |
Below is average latency of generating an image of size 512x512 using NVIDIA A100-SXM4-80GB GPU:
|
42 |
|
43 |
-
| Engine | Batch Size | Steps |
|
44 |
-
|-------------|------------|------ |
|
45 |
-
| Static | 1 | 1 |
|
46 |
-
| Static | 4 | 1 |
|
47 |
-
| Static | 1 | 4 |
|
48 |
-
| Static | 4 | 4 |
|
49 |
|
50 |
|
51 |
Static means the engine is built for the given batch size and image size combination, and CUDA graph is used to speed up.
|
|
|
14 |
- text-to-image
|
15 |
---
|
16 |
|
17 |
+
# Stable Diffusion XL Turbo for ONNX Runtime CUDA
|
18 |
|
19 |
## Introduction
|
20 |
|
21 |
+
This repository hosts the optimized ONNX models of **SD Turbo** to accelerate inference with ONNX Runtime CUDA execution provider for Nvidia GPUs. It cannot run in other providers like CPU and DirectML.
|
22 |
|
23 |
The models are generated by [Olive](https://github.com/microsoft/Olive/tree/main/examples/stable_diffusion) with command like the following:
|
24 |
```
|
|
|
34 |
- **License:** [STABILITY AI NON-COMMERCIAL RESEARCH COMMUNITY LICENSE](https://huggingface.co/stabilityai/sd-turbo/blob/main/LICENSE)
|
35 |
- **Model Description:** This is a conversion of the [SD-Turbo](https://huggingface.co/stabilityai/sd-turbo) model for [ONNX Runtime](https://github.com/microsoft/onnxruntime) inference with CUDA execution provider.
|
36 |
|
37 |
+
## Performance
|
38 |
|
39 |
#### Latency
|
40 |
|
41 |
Below is average latency of generating an image of size 512x512 using NVIDIA A100-SXM4-80GB GPU:
|
42 |
|
43 |
+
| Engine | Batch Size | Steps | ONNX Runtime CUDA |
|
44 |
+
|-------------|------------|------ | ----------------- |
|
45 |
+
| Static | 1 | 1 | 38.2 ms |
|
46 |
+
| Static | 4 | 1 | 120.2 ms |
|
47 |
+
| Static | 1 | 4 | 68.7 ms |
|
48 |
+
| Static | 4 | 4 | 192.6 ms |
|
49 |
|
50 |
|
51 |
Static means the engine is built for the given batch size and image size combination, and CUDA graph is used to speed up.
|