From 85ea9ceff7d785bf53608edd71804512663a6ab8 Mon Sep 17 00:00:00 2001
From: niuliling123 <51102941+niuliling123@users.noreply.github.com>
Date: Wed, 27 Apr 2022 14:37:04 +0800
Subject: [PATCH] fix autotune (#42249)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

fix autotune in reader.py
bug原因：创建dataloader的使用直接使用原始dataloader 等于的方式: new_dataloader = old_dataloader，属于深拷贝，导致后续修改new_dataloader的dataset用于num_workers选择时，会造成old_dataloader的dataset同步被修改。导致模型运行的dataset与实际dataset不相等
---
 python/paddle/fluid/reader.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/python/paddle/fluid/reader.py b/python/paddle/fluid/reader.py
index 841c58821d7..3ea3af9ed1c 100644
--- a/python/paddle/fluid/reader.py
+++ b/python/paddle/fluid/reader.py
@@ -19,6 +19,7 @@ import numpy as np
 import threading
 import paddle
 import time
+import copy
 
 from .framework import Program, Variable, program_guard, default_main_program, default_startup_program, _non_static_mode, cpu_places, _current_expected_place, _in_eager_without_dygraph_check
 from .executor import global_scope
@@ -214,7 +215,7 @@ class AuToTune(object):
         return sub_dataset
 
     def get_autotune_loader(self):
-        loader = self.loader
+        loader = copy.copy(self.loader)
         batch_size = self.loader.batch_sampler.batch_size
         if isinstance(self.loader.batch_sampler,
                       paddle.io.DistributedBatchSampler):
-- 
GitLab