zhangyiss/mlx-examples
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx-examples.git synced 2025-10-24 06:28:07 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Updated CIFAR-10 ResNet example to use BatchNorm instead of LayerNorm (#257)

Browse Source 
* replaced nn.LayerNorm by nn.BatchNorm

* mlx>=0.0.8 required

* updated default to 30 epochs instead of 100

* updated README after adding BatchNorm

* requires mlx>=0.0.9

* updated README.md with results for mlx-0.0.9
This commit is contained in:
Markus Enzweiler
2024-01-12 14:43:11 +01:00
committed by GitHub
parent 6217d7acd0
commit 2b61d9deb6
4 changed files with 10 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
cifar/README.md
View File

@@ -36,16 +36,15 @@ python main.py --help
## Results
After training with the default `resnet20` architecture for 100 epochs, you
After training with the default `resnet20` architecture for 30 epochs, you
should see the following results:
```
Epoch: 99 | avg. Train loss 0.320 | avg. Train acc 0.888 | Throughput: 416.77 images/sec
Epoch: 99 | Test acc 0.807
Epoch: 29 | avg. Train loss 0.294 | avg. Train acc 0.897 | Throughput: 270.81 images/sec
Epoch: 29 | Test acc 0.841
```
Note this was run on an M1 Macbook Pro with 16GB RAM.
At the time of writing, `mlx` doesn't have built-in learning rate schedules,
or a `BatchNorm` layer. We intend to update this example once these features
are added.
At the time of writing, `mlx` doesn't have built-in learning rate schedules.
We intend to update this example once these features are added.