apli

Thanks,@junshi15 I think I get what you mean. Anyway,with more executors,I could set bigger batch size(the actual batch size = batch size * num of executors) to make full use...

Another question, If I have two executers(1 gpu per executor), the gpu of one is idle and another is busy.Does the time cost of training depends mainly on the training...

What's the main factor that affect the communication, the bandwith?

Hi,@arundasan91, I was following the steps in https://github.com/yahoo/CaffeOnSpark/wiki/GetStarted_yarn. I've looked at the post you give, which tells: the spark jar containing the DeveloperApi class was missing from the classpath and...

@junshi15 , Thanks! I tried it with hadoop 2.7.1 ,spark 2.0.0 and java 1.8.0_121, and caffe-grid was built failed: - CocoTest *** FAILED *** org.apache.spark.SparkException: Job aborted due to stage...

@arundasan91 , Thanks! The failed test still existed after I udpate to java to Openjdk 1.8.0_72. 17/03/14 01:34:48 INFO handler.ContextHandler: Started o.s.j.s.ServletContextHandler@4a86b344{/metrics/json,null,AVAILABLE} testTrain (__main__.PythonApiTest) ... java.lang.AssertionError: assertion failed: unsafe symbol...

@junshi15, The failed Test is familiar with the issue #184. And the build passed after I commented the failed testcase in the Makefile.