def test(args, net, loss, opt, dataset, checkpoints):
model = KungFuModel(
net,
loss_fn=loss,
optimizer=opt,
# loss_scale_manager=loss_scale,
metrics={'acc'},
amp_level="O2",
keep_batchnorm_fp32=False,
)
results = []
for ckpt_name in checkpoints:
logical_step = parse_logical_step(ckpt_name)
acc = test_checkpoint(net, model, dataset, ckpt_name)
results.append((logical_step, acc))
msg = "%s, accuracy: %s" % (ckpt_name, acc)
print(msg)
filename = 'plot/lbs-%d+dbs-%d.txt' % (args.logical_batch_size,
args.device_batch_size)
with open(filename, 'w') as f:
for step, acc in results:
# msg = "%s, accuracy: %s" % (ckpt_name, acc)
msg = '%d %s' % (step, acc)
f.write(msg + '\n')
print('saved to %s' % (filename))
def test(args, net, loss, opt, dataset, checkpoints):
metrics = [
'top_1_accuracy',
'top_5_accuracy',
]
model = KungFuModel(
net,
loss_fn=loss,
optimizer=opt,
metrics=set(metrics),
amp_level="O2",
# [ERROR] DEVICE(8546,python3.7):2021-02-27-01:12:19.225.728 [mindspore/ccsrc/runtime/device/gpu/kernel_info_setter.cc:118] SelectAkgKernel] Not find op[BatchNorm] in akg
# [ERROR] DEVICE(8546,python3.7):2021-02-27-01:12:19.225.791 [mindspore/ccsrc/runtime/device/gpu/kernel_info_setter.cc:322] PrintUnsupportedTypeException] Select GPU kernel op[BatchNorm] fail! Incompatible data type!
# The supported data types are in[float32 float32 float32 float32 float32], out[float32 float32 float32 float32 float32]; in[float16 float32 float32 float32 float32], out[float16 float32 float32 float32 float32]; , but get in [float16 float16 float16 float16 float16 ] out [float16 float16 float16 float16 float16 ]
# keep_batchnorm_fp32=False,
)
results = []
for ckpt_name in checkpoints:
epoch, step = parse_logical_step(ckpt_name)
result = test_checkpoint(net, model, dataset, ckpt_name)
results.append((epoch, step, result))
msg = "%s, %s: %s" % (ckpt_name, metrics[0], result[metrics[0]])
print(msg)
filename = get_eval_result_filename(args)
with open(filename, 'w') as f:
for epoch, step, result in sorted(results):
acc = result[metrics[0]]
msg = '%d %d %s' % (epoch, step, acc)
f.write(msg + '\n')
print('%d points saved to %s' % (len(results), filename))
def train(args, net, loss, opt, dataset):
model = KungFuModel(
net,
loss_fn=loss,
optimizer=opt,
# loss_scale_manager=loss_scale,
metrics={'acc'},
amp_level="O2",
keep_batchnorm_fp32=False,
)
model.train(
args,
epoch=args.epochs,
train_dataset=dataset,
callbacks=build_callbacks(args),
dataset_sink_mode=False,
)