add ips print for language_model (#4887)

* fix ptb_dy time print for benchmark, test=develop

* add ips print for language_model, test=develop

PaddleNLP/language_model/train.py

@@ -312,6 +312,7 @@ def main():
         total_loss = 0
         iters = 0
         batch_cost_avg = TimeCostAverage()
+        reader_cost_avg = TimeCostAverage()
 
         init_hidden, init_cell = generate_init_data()
         batch_start_time = time.time()
@@ -323,6 +324,8 @@ def main():
                 epoch_id=epoch_id,
                 with_lr=True,
                 device_count=device_count)
+            reader_time = time.time() - batch_start_time
+            reader_cost_avg.record(reader_time)
             fetch_outs = exe.run(train_program,
                                  feed=input_data_feed,
                                  fetch_list=[
@@ -343,10 +346,13 @@ def main():
             if batch_id > 0 and batch_id % log_interval == 0:
                 ppl = np.exp(total_loss / iters)
                 print(
-                    "-- Epoch:[%d]; Batch:[%d]; Time: %.5f s; ppl: %.5f, lr: %.5f"
-                    % (epoch_id, batch_id, batch_cost_avg.get_average(), ppl[0],
-                       lr[0]))
+                    "-- Epoch:[%d]; Batch:[%d]; ppl: %.5f, lr: %.5f, batch_cost: %.5f sec, reader_cost: %.5f sec, ips: %.5f words/sec"
+                    % (epoch_id, batch_id, ppl[0], lr[0],
+                       batch_cost_avg.get_average(),
+                       reader_cost_avg.get_average(),
+                       config.batch_size / batch_cost_avg.get_average()))
                 batch_cost_avg.reset()
+                reader_cost_avg.reset()
 
             # profiler tools for benchmark
             if args.profile and batch_id == log_interval: