update ONNX-TensorRT descriptions
Browse files- README.md +2 -2
- configs/metadata.json +3 -2
- docs/README.md +2 -2
README.md
CHANGED
@@ -75,7 +75,7 @@ Accuracy was used for evaluating the performance of the model. This model achiev
|
|
75 |
![A graph showing the validation accuracy over 25 epochs.](https://developer.download.nvidia.com/assets/Clara/Images/monai_endoscopic_inbody_classification_val_accuracy_v2.png)
|
76 |
|
77 |
#### TensorRT speedup
|
78 |
-
The `endoscopic_inbody_classification` bundle supports
|
79 |
|
80 |
| method | torch_fp32(ms) | torch_amp(ms) | trt_fp32(ms) | trt_fp16(ms) | speedup amp | speedup fp32 | speedup fp16 | amp vs fp16|
|
81 |
| :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
|
@@ -90,7 +90,7 @@ Where:
|
|
90 |
- `speedup amp`, `speedup fp32` and `speedup fp16` are the speedup ratios of corresponding models versus the PyTorch float32 model
|
91 |
- `amp vs fp16` is the speedup ratio between the PyTorch amp model and the TensorRT float16 based model.
|
92 |
|
93 |
-
Currently,
|
94 |
|
95 |
This result is benchmarked under:
|
96 |
- TensorRT: 8.5.3+cuda11.8
|
|
|
75 |
![A graph showing the validation accuracy over 25 epochs.](https://developer.download.nvidia.com/assets/Clara/Images/monai_endoscopic_inbody_classification_val_accuracy_v2.png)
|
76 |
|
77 |
#### TensorRT speedup
|
78 |
+
The `endoscopic_inbody_classification` bundle supports acceleration with TensorRT through the ONNX-TensorRT method. The table below displays the speedup ratios observed on an A100 80G GPU.
|
79 |
|
80 |
| method | torch_fp32(ms) | torch_amp(ms) | trt_fp32(ms) | trt_fp16(ms) | speedup amp | speedup fp32 | speedup fp16 | amp vs fp16|
|
81 |
| :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
|
|
|
90 |
- `speedup amp`, `speedup fp32` and `speedup fp16` are the speedup ratios of corresponding models versus the PyTorch float32 model
|
91 |
- `amp vs fp16` is the speedup ratio between the PyTorch amp model and the TensorRT float16 based model.
|
92 |
|
93 |
+
Currently, the only available method to accelerate this model is through ONNX-TensorRT. However, the Torch-TensorRT method is under development and will be available in the near future.
|
94 |
|
95 |
This result is benchmarked under:
|
96 |
- TensorRT: 8.5.3+cuda11.8
|
configs/metadata.json
CHANGED
@@ -1,7 +1,8 @@
|
|
1 |
{
|
2 |
"schema": "https://github.com/Project-MONAI/MONAI-extra-test-data/releases/download/0.8.1/meta_schema_20220324.json",
|
3 |
-
"version": "0.4.
|
4 |
"changelog": {
|
|
|
5 |
"0.4.1": "update the model weights with the deterministic training",
|
6 |
"0.4.0": "add the ONNX-TensorRT way of model conversion",
|
7 |
"0.3.9": "fix mgpu finalize issue",
|
@@ -20,7 +21,7 @@
|
|
20 |
"0.1.0": "complete the first version model package",
|
21 |
"0.0.1": "initialize the model package structure"
|
22 |
},
|
23 |
-
"monai_version": "1.2.
|
24 |
"pytorch_version": "1.13.1",
|
25 |
"numpy_version": "1.22.2",
|
26 |
"optional_packages_version": {
|
|
|
1 |
{
|
2 |
"schema": "https://github.com/Project-MONAI/MONAI-extra-test-data/releases/download/0.8.1/meta_schema_20220324.json",
|
3 |
+
"version": "0.4.2",
|
4 |
"changelog": {
|
5 |
+
"0.4.2": "update ONNX-TensorRT descriptions",
|
6 |
"0.4.1": "update the model weights with the deterministic training",
|
7 |
"0.4.0": "add the ONNX-TensorRT way of model conversion",
|
8 |
"0.3.9": "fix mgpu finalize issue",
|
|
|
21 |
"0.1.0": "complete the first version model package",
|
22 |
"0.0.1": "initialize the model package structure"
|
23 |
},
|
24 |
+
"monai_version": "1.2.0rc5",
|
25 |
"pytorch_version": "1.13.1",
|
26 |
"numpy_version": "1.22.2",
|
27 |
"optional_packages_version": {
|
docs/README.md
CHANGED
@@ -68,7 +68,7 @@ Accuracy was used for evaluating the performance of the model. This model achiev
|
|
68 |
![A graph showing the validation accuracy over 25 epochs.](https://developer.download.nvidia.com/assets/Clara/Images/monai_endoscopic_inbody_classification_val_accuracy_v2.png)
|
69 |
|
70 |
#### TensorRT speedup
|
71 |
-
The `endoscopic_inbody_classification` bundle supports
|
72 |
|
73 |
| method | torch_fp32(ms) | torch_amp(ms) | trt_fp32(ms) | trt_fp16(ms) | speedup amp | speedup fp32 | speedup fp16 | amp vs fp16|
|
74 |
| :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
|
@@ -83,7 +83,7 @@ Where:
|
|
83 |
- `speedup amp`, `speedup fp32` and `speedup fp16` are the speedup ratios of corresponding models versus the PyTorch float32 model
|
84 |
- `amp vs fp16` is the speedup ratio between the PyTorch amp model and the TensorRT float16 based model.
|
85 |
|
86 |
-
Currently,
|
87 |
|
88 |
This result is benchmarked under:
|
89 |
- TensorRT: 8.5.3+cuda11.8
|
|
|
68 |
![A graph showing the validation accuracy over 25 epochs.](https://developer.download.nvidia.com/assets/Clara/Images/monai_endoscopic_inbody_classification_val_accuracy_v2.png)
|
69 |
|
70 |
#### TensorRT speedup
|
71 |
+
The `endoscopic_inbody_classification` bundle supports acceleration with TensorRT through the ONNX-TensorRT method. The table below displays the speedup ratios observed on an A100 80G GPU.
|
72 |
|
73 |
| method | torch_fp32(ms) | torch_amp(ms) | trt_fp32(ms) | trt_fp16(ms) | speedup amp | speedup fp32 | speedup fp16 | amp vs fp16|
|
74 |
| :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
|
|
|
83 |
- `speedup amp`, `speedup fp32` and `speedup fp16` are the speedup ratios of corresponding models versus the PyTorch float32 model
|
84 |
- `amp vs fp16` is the speedup ratio between the PyTorch amp model and the TensorRT float16 based model.
|
85 |
|
86 |
+
Currently, the only available method to accelerate this model is through ONNX-TensorRT. However, the Torch-TensorRT method is under development and will be available in the near future.
|
87 |
|
88 |
This result is benchmarked under:
|
89 |
- TensorRT: 8.5.3+cuda11.8
|