提交 5250ca8c 编写于 作者: T tangwei12

bug fux

上级 459690ae
......@@ -43,7 +43,7 @@ class CheckpointNotifyOp : public framework::OperatorBase {
detail::RPCClient::GetInstance<RPCCLIENT_T>();
for (size_t i = 0; i < epmap.size(); i++) {
VLOG(3) << "sending " << dir <<" to " << epmap[i] << " to checkpoint notify ... ";
auto serial_looku_table = string::Sprintf("%s/%s.%d", dir, lookup_table_name, i);
auto serial_looku_table = string::Sprintf("%s/%s_%d", dir, lookup_table_name, i);
rpc_client->AsyncCheckpointNotify(epmap[i], serial_looku_table);
}
rpc_client->Wait();
......
......@@ -462,7 +462,6 @@ CHECKPOINT_PREFIX = "checkpoint"
MODEL_DIR = "__model__"
LOOKUP_TABLE_DIR = "__lookup_table__"
TRAINER_PREFIX = "trainer"
PSERVER_PREFIX = "pserver"
CHECKPOINT_SEPARATOR = "_"
......@@ -577,8 +576,7 @@ def load_persist_vars_without_grad(executor,
def load_lookup_table_vars(executor, dirname, pserver_id, table_name):
lookup_table_dir = os.path.join(dirname, LOOKUP_TABLE_DIR)
table_file = table_name + CHECKPOINT_SEPARATOR + PSERVER_PREFIX + CHECKPOINT_SEPARATOR + str(
pserver_id)
table_file = table_name + CHECKPOINT_SEPARATOR + str(pserver_id)
load_vars(executor, lookup_table_dir, vars=table_name, filename=table_file)
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册