Created by: HugoLian

各位大神，我有个一paddle训练，网络结构类似 book教程中的系统推荐，模型训练到不同的pass之后随机报错退出，退出时报错代码为：

I0728 00:45:40.022342  1868 Stat.cpp:132] Stat=BackwardTimer                  TID=1875   total=101.164    avg=0.046      max=0.461      min=0          count=2160      
                                          Stat=BackwardTimer                  TID=1876   total=105.213    avg=0.048      max=0.739      min=0          count=2160      
                                          Stat=BackwardTimer                  TID=1871   total=110.272    avg=0.051      max=0.553      min=0          count=2160      
                                          Stat=BackwardTimer                  TID=1877   total=110.925    avg=0.051      max=0.546      min=0          count=2160      
                                          Stat=BackwardTimer                  TID=1872   total=105.79     avg=0.048      max=0.673      min=0          count=2160      
                                          Stat=BackwardTimer                  TID=1874   total=107.7      avg=0.049      max=0.536      min=0          count=2160      
                                          Stat=BackwardTimer                  TID=1873   total=104.855    avg=0.048      max=0.727      min=0          count=2160      
I0728 00:45:40.022374  1868 Stat.cpp:140] ======= BarrierStatSet status ======
I0728 00:45:40.022379  1868 Stat.cpp:153] --------------------------------------------------
I0728 00:46:22.938666  1868 Tester.cpp:127]  Test samples=886436 cost=0.619395 Eval: 
I0728 00:46:22.953542  1868 GradientMachine.cpp:112] Saving parameters to ./output2/pass-00030
I0728 00:46:23.030747  1868 Util.cpp:230] copy trainer_config_feed.py to ./output2/pass-00030
*** Aborted at 1501173992 (unix time) try "date -d @1501173992" if you are using GNU date ***
PC: @     0x7fe550fdac98 (unknown)
*** SIGSEGV (@0x0) received by PID 1868 (TID 0x7fe507fff700) from PID 0; stack trace: ***
    @     0x7fe55123e160 (unknown)
    @     0x7fe550fdac98 (unknown)
/home/iknow/lianjie/paddle/paddle_internal_release_tools/idl/paddle/output/bin/paddle_local: line 109:  1868 Segmentation fault      (core dumped) ${DEBUGGER} $MYDIR/../opt/paddle/bin/paddle_trainer ${@:2}

请问训练过程中随机退出都可能有什么原因呢？这个个train.sh的参数设置有关系么？

cfg=trainer_config_feed.py

paddle train \
  --config=$cfg \
  --save_dir=./output2 \
  --trainer_count=7 \
  --log_period=2000 \
  --num_passes=150 \
  --use_gpu=false \
  --show_parameter_stats_period=500 \
  --test_all_data_in_one_period=1 \
  --dot_period=30 \
  --saving_period=1 \
  --num_gradient_servers=1 \
  2>&1 | tee 'train.log.2'

另外原来好像见过一个输入初始pass，就可以从指定参数基础上开始训练，现在找不到了，paddle是支持这个的吧？