未验证 提交 8fa457f9 编写于 作者: C chengduo 提交者: GitHub

Merge pull request #11296 from typhoonzero/fix_fluid_benchmark_performance_data

fix small bug in caculating speed
...@@ -371,8 +371,7 @@ def train_parallel(avg_loss, infer_prog, optimizer, train_reader, test_reader, ...@@ -371,8 +371,7 @@ def train_parallel(avg_loss, infer_prog, optimizer, train_reader, test_reader,
print("Pass %d, batch %d, loss %s" % print("Pass %d, batch %d, loss %s" %
(pass_id, batch_id, np.array(loss))) (pass_id, batch_id, np.array(loss)))
batch_id += 1 batch_id += 1
if args.use_reader_op:
num_samples = num_samples * args.gpus
print_train_time(start_time, time.time(), num_samples) print_train_time(start_time, time.time(), num_samples)
if not args.no_test and batch_acc: if not args.no_test and batch_acc:
test_acc = test(startup_exe, infer_prog, test_reader, feeder, test_acc = test(startup_exe, infer_prog, test_reader, feeder,
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册