Revert "[MXNET-apache#16795] Byteps-KVStore: Intergrate Byteps into m…

…xnet as new type of kvstore backend (apache#17555)"

This reverts commit c244f9f.

ci/docker/runtime_functions.sh

@@ -1004,7 +1004,6 @@ cd_unittest_ubuntu() {
         # Adding these here as CI doesn't test all CUDA environments
         $python_cmd example/image-classification/test_score.py
         integrationtest_ubuntu_gpu_dist_kvstore
-        integrationtest_ubuntu_gpu_byteps
     fi
 
     if [[ ${mxnet_variant} = *mkl ]]; then
@@ -1352,24 +1351,6 @@ integrationtest_ubuntu_gpu_dist_kvstore() {
     popd
 }
 
-integrationtest_ubuntu_gpu_byteps() {
-    set -ex
-    pushd .
-    export PYTHONPATH=$PWD/python/
-    export BYTEPS_WITHOUT_PYTORCH=1
-    export BYTEPS_WITHOUT_TENSORFLOW=1
-    git clone -b v0.2 https://github.com/bytedance/byteps/ --recursive
-    cd byteps && python3 setup.py install --user && cd -
-
-    export MXNET_STORAGE_FALLBACK_LOG_VERBOSE=0
-    export MXNET_SUBGRAPH_VERBOSE=0
-    export DMLC_LOG_STACK_TRACE_DEPTH=10
-    cd tests/nightly/
-    python3 ../../tools/launch.py -n 1 -s 1 --byteps --env NVIDIA_VISIBLE_DEVICES:0,1 python3 dist_device_sync_kvstore_byteps.py
-    popd
-}
-
-
 test_ubuntu_cpu_python3() {
     set -ex
     pushd .

ci/jenkins/Jenkins_steps.groovy

@@ -1277,20 +1277,6 @@ def test_unix_distributed_kvstore_cpu() {
     }]
 }
 
-def test_unix_byteps_gpu() {
-    return ['byteps tests GPU': {
-      node(NODE_LINUX_GPU) {
-        ws('workspace/it-byteps') {
-          timeout(time: max_time, unit: 'MINUTES') {
-            utils.unpack_and_init('gpu', mx_lib)
-            utils.docker_run('ubuntu_gpu_cu101', 'integrationtest_ubuntu_gpu_byteps', true)
-            utils.publish_test_coverage()
-          }
-        }
-      }
-    }]
-}
-
 def test_unix_distributed_kvstore_gpu() {
     return ['dist-kvstore tests GPU': {
       node(NODE_LINUX_GPU) {

ci/jenkins/Jenkinsfile_edge

@@ -44,7 +44,7 @@ core_logic: {
 
   utils.parallel_stage('Tests', [
     custom_steps.test_qemu_armv7_cpu()
-  ])
+  ]) 
 }
 ,
 failure_handler: {

ci/jenkins/Jenkinsfile_unix_gpu

@@ -60,7 +60,6 @@ core_logic: {
     custom_steps.test_unix_cpp_package_gpu(),
     custom_steps.test_unix_scala_gpu(),
     custom_steps.test_unix_distributed_kvstore_gpu(),
-    custom_steps.test_unix_byteps_gpu(),
     custom_steps.test_static_python_gpu(),
     custom_steps.test_static_python_gpu_cmake(),
     custom_steps.test_unix_python3_gpu_no_tvm_op(),

python/mxnet/kvstore/__init__.py

@@ -22,4 +22,3 @@
 from .kvstore import *
 from .base import *
 from .kvstore_server import *
-from .byteps import *

python/mxnet/kvstore/base.py

@@ -313,8 +313,6 @@ def pushpull(self, key, value, out=None, priority=0):
     def is_capable(capability):
         """Queries if the KVStore type supports certain capability, such as optimizer algorithm,
         gradient compression, sparsity, etc.
-        If the kvstore does not store weights in server part, then no optimizer is supported,
-        this function will return False.
 
         Parameters
         ----------
@@ -429,15 +427,10 @@ def create(name='local'):
     No two updates happen on the same weight at the same time. However, the order is not
     guaranteed.
 
-    ``byteps``: Use byteps as broadcast/pushpull backend.
-    This kind of kvstore doesn't store weights, thus there won't be optimizer in this kvstore server.
-    Byteps doesn't support pure cpu training, so be sure to enable gpu training when using this kvstore.
-
     Parameters
     ----------
-    name : {'local', 'device', 'nccl', 'dist_sync', 'dist_device_sync', 'dist_async', 'horovod', 'byteps'}
+    name : {'local', 'device', 'nccl', 'dist_sync', 'dist_device_sync', 'dist_async', 'horovod'}
         The type of KVStore.
-
     Returns
     -------
     kv : KVStoreBase