update quickstart docs
parent
d909faccad
commit
73d3359842
|
@ -75,9 +75,23 @@ python3 -m paddle.distributed.launch \
|
|||
|
||||
The highest accuracy of the validation set is around 0.415.
|
||||
|
||||
* ** Note**
|
||||
Here, multiple GPUs are used for training. If only one GPU is used, please specify the GPU with the `CUDA_VISIBLE_DEVICES` setting, and specify the GPU with the `--gpus` setting, the same below. For example, to train with only GPU 0:
|
||||
|
||||
```shell
|
||||
export CUDA_VISIBLE_DEVICES=0
|
||||
python3 -m paddle.distributed.launch \
|
||||
--gpus="0" \
|
||||
tools/train.py \
|
||||
-c ./ppcls/configs/quick_start/professional/ResNet50_vd_CIFAR100.yaml \
|
||||
-o Global.output_dir="output_CIFAR" \
|
||||
-o Optimizer.lr.learning_rate=0.01
|
||||
```
|
||||
|
||||
* **Notice**:
|
||||
|
||||
* The GPUs specified in `--gpus` can be a subset of the GPUs specified in `CUDA_VISIBLE_DEVICES`.
|
||||
* Since the initial learning rate and batch-size need to maintain a linear relationship, when training is switched from 4 GPUs to 1 GPU, the total batch-size is reduced to 1/4 of the original, and the learning rate also needs to be reduced to 1/4 of the original, so changed the default learning rate from 0.04 to 0.01.
|
||||
|
||||
* If the number of GPU cards is not 4, the accuracy of the validation set may be different from 0.415. To maintain a comparable accuracy, you need to change the learning rate in the configuration file to `the current learning rate / 4 \* current card number`. The same below.
|
||||
|
||||
<a name="2.1.2"></a>
|
||||
|
||||
|
|
|
@ -75,9 +75,22 @@ python3 -m paddle.distributed.launch \
|
|||
|
||||
验证集的最高准确率为 0.415 左右。
|
||||
|
||||
* **注意**
|
||||
此处的使用了多个 GPU 训练,如果只使用一个 GPU,请将 `CUDA_VISIBLE_DEVICES` 设置指定 GPU,`--gpus`设置指定 GPU,下同。例如,只使用 0 号 GPU 训练:
|
||||
|
||||
* 如果 GPU 卡数不是 4,验证集的准确率可能与 0.415 有差异,若需保持相当的准确率,需要将配置文件中的学习率改为`当前学习率 / 4 \* 当前卡数`。下同。
|
||||
```shell
|
||||
export CUDA_VISIBLE_DEVICES=0
|
||||
python3 -m paddle.distributed.launch \
|
||||
--gpus="0" \
|
||||
tools/train.py \
|
||||
-c ./ppcls/configs/quick_start/professional/ResNet50_vd_CIFAR100.yaml \
|
||||
-o Global.output_dir="output_CIFAR" \
|
||||
-o Optimizer.lr.learning_rate=0.01
|
||||
```
|
||||
|
||||
* **注意**:
|
||||
|
||||
* `--gpus`中指定的 GPU 可以是 `CUDA_VISIBLE_DEVICES` 指定的 GPU 的子集。
|
||||
* 由于初始学习率和 batch-size 需要保持线性关系,所以训练从 4 个 GPU 切换到 1 个 GPU 训练时,总 batch-size 缩减为原来的 1/4,学习率也需要缩减为原来的 1/4,所以改变了默认的学习率从 0.04 到 0.01。
|
||||
|
||||
<a name="2.1.2"></a>
|
||||
|
||||
|
@ -157,7 +170,7 @@ python3 -m paddle.distributed.launch \
|
|||
* **注意**
|
||||
|
||||
* 其他数据增广的配置文件可以参考 `ppcls/configs/ImageNet/DataAugment/` 中的配置文件。
|
||||
* 训练 CIFAR100 的迭代轮数较少,因此进行训练时,验证集的精度指标可能会有 1% 左右的波动。
|
||||
* 训练 CIFAR100 的迭代轮数较少,因此进行训练时,验证集的精度指标可能会有 1% 左右的波动。
|
||||
|
||||
<a name="4"></a>
|
||||
|
||||
|
|
Loading…
Reference in New Issue