This repo is migrated to https://github.com/PaddlePaddle/Paddle/tree/develop/benchmark. Please check it out.
Machine:
-
Server: Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz, 2 Sockets, 20 Cores per socket
-
CPU environment System: Ubuntu 16.04.3 LTS, Docker 17.05.0-ce, build 89658be
-
GPU environment System: Ubuntu 16.04.3 LTS, NVIDIA-Docker 17.05.0-ce, build 89658be NVIDIA Docker image: nvidia/cuda:8.0-cudnn5-devel-ubuntu16.04
PaddlePaddle: 0.11.0(Fluid)
- paddlepaddle/paddle:latest
TensorFlow: 1.4.0
- tensorflow/tensorflow:latest
We selected some classic models, compare the performance and speed with TensorFlow.
train cost | train accuracy | test accuracy | samples/sec | train cost | train accuracy | test accuracy | samples/sec | |
---|---|---|---|---|---|---|---|---|
MNIST CNN | ||||||||
VGG-19 | ||||||||
RESNET-101 | ||||||||
Stacked LSTM |
-
TBD add charts compare here
-
VGG-19 input image size - 3 * 224 * 224, Time: images/second
BatchSize | 64 | 128 | 256 |
---|---|---|---|
PaddlePaddle Fluid | |||
TensorFlow |
-
TBD add charts compare here
-
RESNET-101
BatchSize | 64 | 128 | 256 |
---|---|---|---|
PaddlePaddle Fluid | |||
TensorFlow |
- TBD
add charts here
- Stacked LSTM
BatchSize | 64 | 128 | 256 |
---|---|---|---|
PaddlePaddle Fluid | |||
TensorFlow |
- TBD
add charts here
To validate the Fluid performance on general models, we choose the models in book chapter, compare the performance and speed with Paddle 0.10.0.
train cost | train accuracy | test accuracy | samples/sec | train cost | train accuracy | test accuracy | samples/sec | |
---|---|---|---|---|---|---|---|---|
01.fit_a_line | ||||||||
02.recognize_digits | ||||||||
03.image_classification | ||||||||
04.word2vec | ||||||||
05.recommender_system | ||||||||
06.understand_sentiment | ||||||||
07.label_semantic_roles | ||||||||
08.machine_translation |