add gru4rec ce

fluid/PaddleRec/gru4rec/.run_ce.sh

@@ -4,13 +4,19 @@ export MKL_NUM_THREADS=1
 export OMP_NUM_THREADS=1
 
 
+export CPU_NUM=1
+export NUM_THREADS=1
+
+FLAGS_benchmark=true  python train.py --train_dir train_big_data --vocab_path vocab_big.txt --use_cuda 0 --batch_size 500 --model_dir model_output --pass_num 2 --enable_ce --step_num 10 | python _ce.py
+
+
 cudaid=${tagspace:=0} # use 0-th card as default
 export CUDA_VISIBLE_DEVICES=$cudaid
 
-FLAGS_benchmark=true  python train.py --train_dir train_big_data --vocab_path vocab_big.txt --use_cuda 1 --batch_size 500 --model_dir model_output --pass_num 2 --enable_ce | python _ce.py
+FLAGS_benchmark=true  python train.py --train_dir train_big_data --vocab_path vocab_big.txt --use_cuda 1 --batch_size 500 --model_dir model_output --pass_num 2 --enable_ce --step_num 1000 | python _ce.py
 
 
 cudaid=${tagspace_4:=0,1,2,3} # use 0-th card as default
 export CUDA_VISIBLE_DEVICES=$cudaid
 
-FLAGS_benchmark=true  python train.py --train_dir train_big_data --vocab_path vocab_big.txt --use_cuda 1 --parallel 1 --num_devices 2 --batch_size 500 --model_dir model_output --pass_num 2 --enable_ce | python _ce.py
+FLAGS_benchmark=true  python train.py --train_dir train_big_data --vocab_path vocab_big.txt --use_cuda 1 --parallel 1 --num_devices 2 --batch_size 500 --model_dir model_output --pass_num 2 --enable_ce --step_num 1000 | python _ce.py

fluid/PaddleRec/gru4rec/_ce.py

@@ -8,12 +8,16 @@ from kpi import DurationKpi
 from kpi import AccKpi
 
 
+each_pass_duration_cpu1_thread1_kpi = DurationKpi('each_pass_duration_cpu1_thread1', 0.08, 0, actived=True)
+train_ppl_cpu1_thread1_kpi = CostKpi('train_ppl_cpu1_thread1', 0.08, 0)
 each_pass_duration_gpu1_kpi = DurationKpi('each_pass_duration_gpu1', 0.08, 0, actived=True)
 train_ppl_gpu1_kpi = CostKpi('train_ppl_gpu1', 0.08, 0)
 each_pass_duration_gpu4_kpi = DurationKpi('each_pass_duration_gpu4', 0.08, 0, actived=True)
 train_ppl_gpu4_kpi = CostKpi('train_ppl_gpu4', 0.08, 0)
 
 tracking_kpis = [
+        each_pass_duration_cpu1_thread1_kpi,
+        train_ppl_cpu1_thread1_kpi,
         each_pass_duration_gpu1_kpi,
         train_ppl_gpu1_kpi,
         each_pass_duration_gpu4_kpi,

fluid/PaddleRec/gru4rec/train.py

@@ -40,6 +40,8 @@ def parse_args():
         '--base_lr', type=float, default=0.01, help='learning rate')
     parser.add_argument(
         '--num_devices', type=int, default=1, help='Number of GPU devices')
+    parser.add_argument(
+        '--step_num', type=int, default=1000, help='Number of steps')
     parser.add_argument(
         '--enable_ce',
         action='store_true',
@@ -116,7 +118,7 @@ def train():
             ce_info.append(newest_ppl)
             if i % args.print_batch == 0:
                 print("step:%d ppl:%.3f" % (i, newest_ppl))
-            if args.enable_ce and i > 1000:
+            if args.enable_ce and i > args.step_num:
                 break
 
         t1 = time.time()