modify some modelzoo testcase threshold

This commit is contained in:
anzhengqi 2021-05-13 14:24:51 +08:00
parent 9c654c6c83
commit 8ea100172e
6 changed files with 26 additions and 26 deletions

View File

@ -29,24 +29,24 @@ def test_BGCF_amazon_beauty():
utils.copy_files(model_path, cur_path, model_name)
cur_model_path = os.path.join(cur_path, model_name)
old_list = ["--datapath=../data_mr"]
new_list = ["--datapath={}".format(os.path.join(utils.data_root, "amazon_beauty/mindrecord_train"))]
utils.exec_sed_command(old_list, new_list, os.path.join(cur_model_path, "scripts/run_train_ascend.sh"))
old_list = ["default=600,"]
new_list = ["default=50,"]
utils.exec_sed_command(old_list, new_list, os.path.join(cur_model_path, "src/config.py"))
old_list = ["context.set_context(device_id=int(parser.device))"]
new_list = ["context.set_context()"]
old_list = ["context.set_context(device_id=int(parser.device))",
"save_checkpoint("]
new_list = ["context.set_context()",
"pass \\# save_checkpoint("]
utils.exec_sed_command(old_list, new_list, os.path.join(cur_model_path, "train.py"))
exec_network_shell = "cd {}/scripts; bash run_train_ascend.sh".format(model_name)
data_path = os.path.join(utils.data_root, "amazon_beauty/mindrecord_train")
exec_network_shell = "cd {}; python train.py --datapath={} &> log"\
.format(model_name, data_path)
print("train process is running...")
ret = os.system(exec_network_shell)
assert ret == 0
print("train process finished.")
cmd = "ps -ef|grep python |grep train.py|grep amazon_beauty|grep -v grep"
ret = utils.process_check(300, cmd)
assert ret
log_file = os.path.join(cur_model_path, "scripts/train/log")
log_file = os.path.join(cur_model_path, "log")
pattern1 = r"loss ([\d\.\+]+)\,"
loss_list = utils.parse_log_file(pattern1, log_file)
loss_list = loss_list[-5:]

View File

@ -28,8 +28,8 @@ def test_center_net():
model_name = "centernet"
utils.copy_files(model_path, cur_path, model_name)
cur_model_path = os.path.join(cur_path, model_name)
old_list = ['new_repeat_count, dataset']
new_list = ['5, dataset']
old_list = ['new_repeat_count, dataset', 'args_opt.data_sink_steps']
new_list = ['5, dataset', '20']
utils.exec_sed_command(old_list, new_list, os.path.join(cur_model_path, "train.py"))
old_list = ["device_ips = {}", "device_ip.strip()",
"rank_size = 0", "this_server = server",
@ -47,16 +47,16 @@ def test_center_net():
.format(dataset_path, utils.rank_table_path)
os.system(exec_network_shell)
cmd = "ps -ef |grep train.py | grep coco | grep -v grep"
ret = utils.process_check(100, cmd)
ret = utils.process_check(120, cmd)
assert ret
log_file = os.path.join(cur_model_path, "LOG{}/training_log.txt")
for i in range(8):
per_step_time = utils.get_perf_data(log_file.format(i))
assert per_step_time < 385
assert per_step_time < 435
loss_list = []
for i in range(8):
loss_cmd = "grep -nr \"outputs are\" {} | awk '{{print $14}}' | awk -F\")\" '{{print $1}}'"\
.format(log_file.format(i))
loss = utils.get_loss_data_list(log_file.format(i), cmd=loss_cmd)
loss_list.append(loss[-1])
assert sum(loss_list) / len(loss_list) < 35.0
assert sum(loss_list) / len(loss_list) < 58.8

View File

@ -38,7 +38,7 @@ def test_DeeplabV3_voc2007():
old_list = ['model.train(args.train_epochs',
'callbacks=cbs']
new_list = ['model.train(70',
new_list = ['model.train(30',
'callbacks=cbs, sink_size=2']
utils.exec_sed_command(old_list, new_list, os.path.join(cur_model_path, "train.py"))
@ -54,10 +54,10 @@ def test_DeeplabV3_voc2007():
for i in range(8):
per_step_time = utils.get_perf_data(log_file.format(i))
print("per_step_time is", per_step_time)
assert per_step_time < 530.0
assert per_step_time < 585.0
loss_list = []
for i in range(8):
loss = utils.get_loss_data_list(log_file.format(i))
print("loss is", loss[-1])
loss_list.append(loss[-1])
assert sum(loss_list) / len(loss_list) < 2.5
assert sum(loss_list) / len(loss_list) < 3.5

View File

@ -30,7 +30,7 @@ def test_gnmtv2_WMT_English_German():
cur_model_path = os.path.join(cur_path, model_name)
old_list = ['dataset_sink_mode=config.dataset_sink_mode']
new_list = ['dataset_sink_mode=config.dataset_sink_mode, sink_size=100']
new_list = ['dataset_sink_mode=config.dataset_sink_mode, sink_size=25']
utils.exec_sed_command(old_list, new_list, os.path.join(cur_model_path, "train.py"))
old_list = ['"epochs": 6,']
new_list = ['"epochs": 4,']
@ -50,7 +50,7 @@ def test_gnmtv2_WMT_English_German():
for i in range(8):
per_step_time = utils.get_perf_data(log_file.format(i, i))
print("per_step_time is", per_step_time)
assert per_step_time < 270.0
assert per_step_time < 330.0
log_file = os.path.join(cur_model_path, "scripts/device{}/loss.log")
loss_list = []
@ -61,4 +61,4 @@ def test_gnmtv2_WMT_English_German():
loss_list.append(loss[-1])
print("loss_list is", loss_list)
print(sum(loss_list) / len(loss_list))
assert sum(loss_list) / len(loss_list) < 120
assert sum(loss_list) / len(loss_list) < 260

View File

@ -41,7 +41,7 @@ def test_SSD_mobilenet_v1_fpn_coco2017():
new_list = ["ssd_mobilenet_v1_fpn"]
utils.exec_sed_command(old_list, new_list, os.path.join(cur_model_path, "src/config.py"))
old_list = ["args_opt.epoch_size", "dataset_sink_mode=dataset_sink_mode"]
new_list = ["5", "dataset_sink_mode=dataset_sink_mode, sink_size=100"]
new_list = ["5", "dataset_sink_mode=dataset_sink_mode, sink_size=20"]
utils.exec_sed_command(old_list, new_list, os.path.join(cur_model_path, "train.py"))
exec_network_shell = "cd {0}; sh -x scripts/run_distribute_train.sh 8 {1} 0.2 coco {2}"\
@ -55,10 +55,10 @@ def test_SSD_mobilenet_v1_fpn_coco2017():
for i in range(8):
per_step_time = utils.get_perf_data(log_file.format(i))
print("per_step_time is", per_step_time)
assert per_step_time < 545
assert per_step_time < 580
loss_list = []
for i in range(8):
loss = utils.get_loss_data_list(log_file.format(i))
print("loss is", loss[-1])
loss_list.append(loss[-1])
assert sum(loss_list) / len(loss_list) < 2.72
assert sum(loss_list) / len(loss_list) < 4.9

View File

@ -245,6 +245,6 @@ def test_yolov3_darknet_8p():
pattern1 = r", *([\d\.]+) imgs/sec"
pattern2 = r"loss:*([\d\.]+),"
fps_list = utils.parse_log_file(pattern1, train_log_file)[1:]
assert sum(fps_list) / len(fps_list) > 480
assert sum(fps_list) / len(fps_list) > 370
loss_list = utils.parse_log_file(pattern2, train_log_file)
assert loss_list[-1] < 280