Skip to content

M
models

PaddlePaddle / models
大约 1 年 前同步成功

通知 222
Star 6828
Fork 2962
M
models
未验证 提交 5e41760f 编写于 作者: Z Zeng Jinle 提交者： GitHub
浏览文件
操作

remove memory optimize usage in scripts, test=develop (#3242)

上级 a0e73bf7
显示空白变更内容
内联 并排
Showing with 0 addition and 65 deletion
PaddleCV/image_classification/README.md
浏览文件 @ 5e41760f
......@@ -68,7 +68,6 @@ python train.py \
--class_dim=1000 \
--image_shape=3,224,224 \
--model_save_dir=output/ \
--with_mem_opt=False \
--with_inplace=True \
--lr_strategy=piecewise_decay \
--lr=0.1
......@@ -83,7 +82,6 @@ python train.py \
* **class_dim**: the class number of the classification task. Default: 1000.
* **image_shape**: input size of the network. Default: "3,224,224".
* **model_save_dir**: the directory to save trained model. Default: "output".
* **with_mem_opt**: whether to use memory optimization or not. Default: False.
* **with_inplace**: whether to use inplace memory optimization or not. Default: True.
* **lr_strategy**: learning rate changing strategy. Default: "piecewise_decay".
* **lr**: initialized learning rate. Default: 0.1.
......@@ -154,8 +152,6 @@ Note: Add and adjust other parameters accroding to specific models and tasks.
You may add `--fp16=1` to start train using mixed precisioin training, which the training process will use float16 and the output model ("master" parameters) is saved as float32. You also may need to pass `--scale_loss` to overcome accuracy issues, usually `--scale_loss=8.0` will do.
Note that currently `--fp16` can not use together with `--with_mem_opt`, so pass `--with_mem_opt=0` to disable memory optimization pass.
### CE
CE is only for internal testing, don't have to set it.
......
PaddleCV/image_classification/README_cn.md
浏览文件 @ 5e41760f
......@@ -64,7 +64,6 @@ python train.py \
--class_dim=1000 \
--image_shape=3,224,224 \
--model_save_dir=output/ \
--with_mem_opt=False \
--with_inplace=True \
--lr_strategy=piecewise_decay \
--lr=0.1
......@@ -79,7 +78,6 @@ python train.py \
* **class_dim**: 类别数，默认值: 1000
* **image_shape**: 图片大小，默认值: "3,224,224"
* **model_save_dir**: 模型存储路径，默认值: "output/"
* **with_mem_opt**: 是否开启显存优化，默认值: False
* **with_inplace**: 是否开启inplace显存优化，默认值: True
* **lr_strategy**: 学习率变化策略，默认值: "piecewise_decay"
* **lr**: 初始学习率，默认值: 0.1
......@@ -142,8 +140,6 @@ python infer.py \
可以通过开启`--fp16=True`启动混合精度训练，这样训练过程会使用float16数据，并输出float32的模型参数（"master"参数）。您可能需要同时传入`--scale_loss`来解决fp16训练的精度问题，通常传入`--scale_loss=8.0`即可。
注意，目前混合精度训练不能和内存优化功能同时使用，所以需要传`--with_mem_opt=False`这个参数来禁用内存优化功能。
### CE测试
注意：CE相关代码仅用于内部测试，enable_ce默认设置False。
......
PaddleCV/image_classification/dist_train/dist_train.py
浏览文件 @ 5e41760f
......@@ -46,7 +46,6 @@ def parse_args():
add_arg('class_dim', int, 1000, "Class number.")
add_arg('image_shape', str, "3,224,224", "input image size")
add_arg('model_save_dir', str, "output", "model save directory")
add_arg('with_mem_opt', bool, False, "Whether to use memory optimization or not.")
add_arg('pretrained_model', str, None, "Whether to use pretrained model.")
add_arg('checkpoint', str, None, "Whether to resume checkpoint.")
add_arg('lr', float, 0.1, "set learning rate.")
......
PaddleCV/image_classification/run.sh
浏览文件 @ 5e41760f
......@@ -7,7 +7,6 @@ python train.py \
--class_dim=1000 \
--image_shape=3,224,224 \
--model_save_dir=output/ \
--with_mem_opt=True \
--lr_strategy=cosine_decay \
--lr=0.1 \
--num_epochs=200 \
......@@ -22,7 +21,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=piecewise_decay \
# --num_epochs=120 \
# --lr=0.01 \
......@@ -39,7 +37,6 @@ python train.py \
# --model_save_dir=output/ \
# --lr=0.02 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --l2_decay=1e-4
#SqueezeNet1_1
......@@ -53,7 +50,6 @@ python train.py \
# --model_save_dir=output/ \
# --lr=0.02 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --l2_decay=1e-4
#VGG11:
......@@ -67,7 +63,6 @@ python train.py \
# --model_save_dir=output/ \
# --lr=0.1 \
# --num_epochs=90 \
# --with_mem_opt=True \
# --l2_decay=2e-4
#VGG13:
......@@ -81,7 +76,6 @@ python train.py \
# --lr=0.01 \
# --num_epochs=90 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --l2_decay=3e-4
#VGG16:
......@@ -95,7 +89,6 @@ python train.py \
# --model_save_dir=output/ \
# --lr=0.01 \
# --num_epochs=90 \
# --with_mem_opt=True \
# --l2_decay=3e-4
#VGG19:
......@@ -108,7 +101,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.01 \
# --num_epochs=90 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=3e-4
......@@ -120,7 +112,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=piecewise_decay \
# --num_epochs=120 \
# --lr=0.1 \
......@@ -134,7 +125,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_decay \
# --num_epochs=240 \
# --lr=0.1 \
......@@ -150,7 +140,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_decay \
# --num_epochs=240 \
# --lr=0.1 \
......@@ -166,7 +155,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_decay \
# --num_epochs=240 \
# --lr=0.1 \
......@@ -180,7 +168,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_decay \
# --num_epochs=240 \
# --lr=0.1 \
......@@ -194,7 +181,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_decay \
# --num_epochs=240 \
# --lr=0.1 \
......@@ -208,7 +194,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_warmup_decay \
# --num_epochs=240 \
# --lr=0.5 \
......@@ -225,7 +210,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_warmup_decay \
# --num_epochs=240 \
# --lr=0.5 \
......@@ -242,7 +226,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_warmup_decay \
# --num_epochs=240 \
# --lr=0.5 \
......@@ -259,7 +242,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_warmup_decay \
# --num_epochs=240 \
# --lr=0.5 \
......@@ -274,7 +256,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_warmup_decay \
# --num_epochs=240 \
# --lr=0.25 \
......@@ -290,7 +271,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_warmup_decay \
# --num_epochs=240 \
# --lr=0.25 \
......@@ -304,7 +284,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_warmup_decay \
# --lr=0.5 \
# --num_epochs=240 \
......@@ -318,7 +297,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=120 \
......@@ -332,7 +310,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=120 \
......@@ -346,7 +323,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=piecewise_decay \
# --num_epochs=120 \
# --lr=0.1 \
......@@ -362,7 +338,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=200 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4 \
......@@ -376,7 +351,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=200 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=7e-5 \
# --use_mixup=True \
......@@ -391,7 +365,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=piecewise_decay \
# --num_epochs=120 \
# --lr=0.1 \
......@@ -407,7 +380,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=200 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4 \
# --use_mixup=True \
......@@ -423,7 +395,6 @@ python train.py \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --lr_strategy=piecewise_decay \
# --with_mem_opt=True \
# --lr=0.1 \
# --num_epochs=120 \
# --l2_decay=1e-4
......@@ -438,7 +409,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=200 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4 \
# --use_mixup=True \
......@@ -455,7 +425,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=200 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4 \
# --use_mixup=True \
......@@ -472,7 +441,6 @@ python train.py \
# --lr_strategy=piecewise_decay \
# --lr=0.1 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4
......@@ -486,7 +454,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=200 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4 \
# --use_mixup=True \
......@@ -503,7 +470,6 @@ python train.py \
# --lr_strategy=piecewise_decay \
# --lr=0.1 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4
......@@ -517,7 +483,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=200 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4 \
# --use_mixup=True \
......@@ -534,7 +499,6 @@ python train.py \
# --lr_strategy=piecewise_decay \
# --lr=0.1 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4
......@@ -548,7 +512,6 @@ python train.py \
# --lr_strategy=piecewise_decay \
# --lr=0.1 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=15e-5
......@@ -562,7 +525,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=200 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4 \
# --use_mixup=True \
......@@ -579,7 +541,6 @@ python train.py \
# --lr_strategy=piecewise_decay \
# --lr=0.1 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4
......@@ -593,7 +554,6 @@ python train.py \
# --lr_strategy=piecewise_decay \
# --lr=0.1 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=18e-5
......@@ -607,7 +567,6 @@ python train.py \
# --lr_strategy=piecewise_decay \
# --lr=0.1 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4
......@@ -621,7 +580,6 @@ python train.py \
# --lr_strategy=piecewise_decay \
# --lr=0.1 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4
......@@ -635,7 +593,6 @@ python train.py \
# --lr_strategy=piecewise_decay \
# --lr=0.1 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4
......@@ -649,7 +606,6 @@ python train.py \
# --lr_strategy=piecewise_decay \
# --lr=0.1 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4
......@@ -663,7 +619,6 @@ python train.py \
# --lr_strategy=piecewise_decay \
# --lr=0.1 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4
......@@ -678,7 +633,6 @@ python train.py \
# --model_save_dir=output/ \
# --lr=0.1 \
# --num_epochs=200 \
# --with_mem_opt=True \
# --l2_decay=1.2e-4
#SE_ResNeXt101_32x4d:
......@@ -692,7 +646,6 @@ python train.py \
# --model_save_dir=output/ \
# --lr=0.1 \
# --num_epochs=200 \
# --with_mem_opt=True \
# --l2_decay=1.5e-5
# SE_154
......@@ -705,7 +658,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=200 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4 \
# --use_mixup=True \
......@@ -720,7 +672,6 @@ python train.py \
# --class_dim=1000 \
# --image_shape=3,224,224 \
# --model_save_dir=output/ \
# --with_mem_opt=True \
# --lr_strategy=cosine_decay \
# --lr=0.01 \
# --num_epochs=200 \
......@@ -736,7 +687,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.045 \
# --num_epochs=120 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4 \
# --resize_short_size=320
......@@ -751,7 +701,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.045 \
# --num_epochs=200 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4 \
# --use_mixup=True \
......@@ -769,7 +718,6 @@ python train.py \
# --lr_strategy=cosine_decay \
# --lr=0.1 \
# --num_epochs=200 \
# --with_mem_opt=True \
# --model_save_dir=output/ \
# --l2_decay=1e-4 \
# --use_mixup=True \
......@@ -787,7 +735,6 @@ python train.py \
# --image_shape=3,224,224 \
# --lr=0.001 \
# --num_epochs=120 \
# --with_mem_opt=False \
# --model_save_dir=output/ \
# --lr_strategy=adam \
# --use_gpu=False
......
PaddleCV/image_classification/train.py
浏览文件 @ 5e41760f
......@@ -443,8 +443,6 @@ def train(args):
use_ngraph = os.getenv('FLAGS_use_ngraph')
if not use_ngraph:
build_strategy = fluid.BuildStrategy()
# memopt may affect GC results
#build_strategy.memory_optimize = args.with_mem_opt
build_strategy.enable_inplace = args.with_inplace
#build_strategy.fuse_all_reduce_ops=1
......
PaddleNLP/Research/ACL2019-JEMT/README.md
浏览文件 @ 5e41760f
......@@ -81,7 +81,6 @@ beta代表发音信息的权重。这表明，即使将绝大部分权重放在
--sort_type pool \
--pool_size 200000 \
--use_py_reader False \
--use_mem_opt False \
--enable_ce False \
--fetch_steps 1 \
pass_num 100 \
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或