From d98480cff5fe2e08fadc79fccd5bce9ab01ed28c Mon Sep 17 00:00:00 2001 From: tangwei12 Date: Fri, 18 May 2018 21:55:33 +0800 Subject: [PATCH] fix serial number --- paddle/fluid/operators/checkpoint_load_op.cc | 4 ++-- python/paddle/fluid/transpiler/distribute_transpiler.py | 3 +++ 2 files changed, 5 insertions(+), 2 deletions(-) diff --git a/paddle/fluid/operators/checkpoint_load_op.cc b/paddle/fluid/operators/checkpoint_load_op.cc index 6c88cbdab0..18871e56c5 100644 --- a/paddle/fluid/operators/checkpoint_load_op.cc +++ b/paddle/fluid/operators/checkpoint_load_op.cc @@ -114,8 +114,8 @@ class CheckpointLoadOp : public framework::OperatorBase { std::string dir = Attr("dir"); std::string serial_num_attr = Attr("Serial"); - PADDLE_ENFORCE(IsNumber(serial_num_attr), - "Checkpoint Serial must be a number"); + VLOG(3) << "CheckpointLoadOp get Attr dir: " << dir; + VLOG(3) << "CheckpointLoadOp get Attr Serial: " << serial_num_attr; std::string serial_var_name = std::string(SERIAL_VAR); auto *serial_var = scope.FindVar(serial_var_name); diff --git a/python/paddle/fluid/transpiler/distribute_transpiler.py b/python/paddle/fluid/transpiler/distribute_transpiler.py index 335dc2342d..bb1e026a6b 100644 --- a/python/paddle/fluid/transpiler/distribute_transpiler.py +++ b/python/paddle/fluid/transpiler/distribute_transpiler.py @@ -654,6 +654,9 @@ class DistributeTranspiler: if os.path.isfile(success_path): return int(cur_dir) + if os.path.isdir(checkpoint_dir): + return "-1" + current_dir = 0 dirs = os.listdir(checkpoint_dir) for cur_dir in dirs: -- GitLab