RetroSearch Browse

Home - News ( United States | United Kingdom | Italy | Germany ) - Football scores

Showing content from https://github.com/Microsoft/multiverso/wiki/Multiverso-Python-Binding-Benchmark below:

Multiverso Python Binding Benchmark · microsoft/Multiverso Wiki · GitHub

Multiverso Python Binding Benchmark

Perform CIFAR-10 classification with residual networks implementation based on Lasagne.

Deep_Residual_Learning_CIFAR-10

Please follow this guide to setup your environment.

Hosts 1 GPU Tesla K40m * 8 CPU Intel(R) Xeon(R) CPU E5-2680 v2 @ 2.80GHz Memory 251GB

Configuration of ~/.theanorc

[global]
device = gpu
floatX = float32

[cuda]
root = /usr/local/cuda-7.5/

[lib]
cnmem = 1

Total epoch 82 Batch size 128 Depth 32 Learning rate change schedule Initialized as 0.1, Changed to 0.01 from epoch 41, to 0.001 from epoch 61 number of parameters in model 464,154

Clarification

An epoch represents all the processes divide all the data equally and go through them once together.
A barrier is used at the end of each epoch.
This experiment doesn't use warm start in ASGD.
The time to load the data is not considered in the time of the experiment.

The results of 3 experiments with different configurations are shown as following.

Short Name # Process(es) #GPU(s) per Process Use multiverso Batch size Initial learning rate Seconds per epoch Best model validation accuracy 1P1G0M 1 1 0 128 0.1 175.4 92.69 % 1P1G1M 1 1 1 128 0.1 194.4 92.53 % 8P1G1M 8 1 1 64 0.05 34.1 92.11 %

RetroSearch is an open source project built by @garambo | Open a GitHub Issue

Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo

HTML: 3.2 | Encoding: UTF-8 | Version: 0.7.4