You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
`dist/kakiage-<version>-py3-none-any.whl`will be generated. The user runs `pip install /path/to/kakiage-<version>-py3-none-any.whl` to install kakiage along with required dependencies (numpy, etc.).
Training MLPs to classify MNIST image datasets in data-parallel distributed training
4
4
5
-
# ビルド
5
+
CIFAR10, CIFAR100 dataset can be also used.
6
+
7
+
# Build
6
8
7
9
```
8
10
npm install
9
11
npm run build
10
12
```
11
13
12
-
# 学習実行
14
+
# Run training
15
+
16
+
Settings are made via environment variables.
17
+
18
+
- MODEL: one of mlp, conv, resnet18. Specify model type.
19
+
- N_CLIENTS: The number of clients participating in the distribution calculation, an integer greater than or equal to 1. If not specified, 1 is assumed to be specified.
20
+
- EPOCH: Number of learning epochs. Default is 2.
21
+
- BATCH_SIZE: Batch size. Total for all clients. Default is 32.
13
22
23
+
Execution is via uvicorn. Command sample (for Mac/Linux):
Open [http://localhost:8081/](http://localhost:8081/) with web browser. If you set `N_CLIENTS`, to run `N_CLIENTS` distributed clients, it must be opened in `N_CLIENTS` browser windows. Note: If three tabs are opened on one window, the computation speed of the tabs not displayed will be reduced.
38
+
39
+
The learned models are output in ONNX format and can be used for inference with WebDNN, ONNX Runtime Web, etc.
0 commit comments