tensorflow - How to execute distributed training where each node has multiple workers -


what command run distributed training on multiple nodes each node has multiple gpus. example in https://github.com/tensorflow/models/tree/master/inception shows case each node has 1 gpu/1 worker. in cluster, each node has 4 gpus should require 4 workers.

i tried following command: on node 0:

bazel-bin/inception/imagenet_distributed_train  --batch_size=32  --data_dir=$home/imagenet-data  --job_name='worker'  --task_id=0  --ps_hosts='ps0.example.com:2222'  --worker_hosts='worker0.example.com:2222,worker0.example.com:2222,worker0.example.com:2222,worker0.example.com:2222,worker1.example.com:2222,worker1.example.com:2222,worker1.example.com:2222,worker1.example.com:2222' & ......  bazel-bin/inception/imagenet_distributed_train  --batch_size=32  --data_dir=$home/imagenet-data  --job_name='worker'  --task_id=3  --ps_hosts='ps0.example.com:2222'  --worker_hosts='worker0.example.com:2222,worker0.example.com:2222,worker0.example.com:2222,worker0.example.com:2222,worker1.example.com:2222,worker1.example.com:2222,worker1.example.com:2222,worker1.example.com:2222'

on node 1:

bazel-bin/inception/imagenet_distributed_train  --batch_size=32  --data_dir=$home/imagenet-data  --job_name='worker'  --task_id=4  --ps_hosts='ps0.example.com:2222'  --worker_hosts='worker0.example.com:2222,worker0.example.com:2222,worker0.example.com:2222,worker0.example.com:2222,worker1.example.com:2222,worker1.example.com:2222,worker1.example.com:2222,worker1.example.com:2222' & ......  bazel-bin/inception/imagenet_distributed_train  --batch_size=32  --data_dir=$home/imagenet-data  --job_name='worker'  --task_id=7  --ps_hosts='ps0.example.com:2222'  --worker_hosts='worker0.example.com:2222,worker0.example.com:2222,worker0.example.com:2222,worker0.example.com:2222,worker1.example.com:2222,worker1.example.com:2222,worker1.example.com:2222,worker1.example.com:2222'

note there & @ end of each command can executed in parallel, has out of gpu memory error.

i tried use 1 worker in each node , each worker uses 4 gpu: on node 0:

bazel-bin/inception/imagenet_distributed_train  --batch_size=32  --data_dir=$home/imagenet-data  --job_name='worker'  --gpus=4 --task_id=0  --ps_hosts='ps0.example.com:2222'  --worker_hosts='worker0.example.com:2222,worker1.example.com:2222'

on node 1:

bazel-bin/inception/imagenet_distributed_train  --batch_size=32  --data_dir=$home/imagenet-data  --job_name='worker'  --gpus=4 --task_id=1  --ps_hosts='ps0.example.com:2222'  --worker_hosts='worker0.example.com:2222,worker1.example.com:2222'

but in end each node uses 1 gpu.

so exact command should use? thanks.


Comments

Post a Comment

Popular posts from this blog

c# - Update a combobox from a presenter (MVP) -

i using mvp in project, new in mvp. have 2 comboboxes. when select option in combobox, combobox should filled new data. this action in presenter. view 'view1' in presenter, , introduced combobox1 , combobox2 properties in 'view1', because need 'datasource', 'displaymember', 'valuemember' , 'refresh()' in method below. but, when using pattern, enough send property like public string combobox2 { { return combobox1.selectedvalue.tosstring(); } } into presenter not whole combobox. how can solve problem? public void onselectedindexchangedcombobox1() { if (view1.combobox1.selectedindex == -1) { return; } datatable dt = tools.getdatatable("a path"); var query = (from o in dt.asenumerable() o.field<string>("afield") == farmerview.combobox1.selectedvalue.tostring() orderby o.field<string>("anotherfield") select...
Read more

How to understand 2 main() functions after using uftrace to profile the C++ program? -

i trying uftrace profile following simple c++ program: #include <iostream> class { public: a() {std::cout << "a created" << std::endl;} ~a() {std::cout << "a destroyed" << std::endl;} }; int main() { a; return 0; } the profile result this: # uftrace a.out created destroyed # duration tid function 2.026 [ 4828] | __cxa_atexit(); [ 4828] | main() { [ 4828] | __static_initialization_and_destruction_0() { 89.397 [ 4828] | std::ios_base::init::init(); 0.768 [ 4828] | __cxa_atexit(); 93.029 [ 4828] | } /* __static_initialization_and_destruction_0 */ 94.425 [ 4828] | } /* main */ [ 4828] | main() { [ 4828] | a::a() { 11.104 [ 4828] | std::operator <<(); 10.825 [ 4828] | std::basic_ostream::operator <<(); 24.514 [ 4828] | } /* a::a */ [ 4828] | a::~a() { 0.978 [ 4828] | ...
Read more

How to put a lock and transaction on table using spring 4 or above using jdbcTemplate and annotations like @Transactional? -

i trying put lock on table while writing on table , if happened in between roll-back . trying convert below code lock table test_g1 read; lock table test_g write; -- begin; start transaction; insert test_g1 values(143); insert test_g values(145); select * test_g1; select * test_g; rollback; select * test_g; unlock tables; how convert above code @transactional spring jdbctemplate code? @transactional(rollbackfor={dataaccessexception.class}) public void test(){ jdbctemplate.execute("insert test1 (id, nam) values (4, 'a')"); throw new dataaccessexception("error") { }; } here trying throw error, insert statement should rollback it's not happening . thanks edit-1 attaching code , doing in jdbcdaoimpl.java , mentioned problem comment above test() . app.java package com.cgiri.javabrains.spring4; import org.springframework.beans.factory.annotation.autowired; import org.springframework.context.applic...
Read more