Skip to content

M
models

PaddlePaddle / models
1 年多 前同步成功

通知 222
Star 6828
Fork 2962
M
models
提交 c7e09860 编写于 作者: Z Zhen Wang
浏览文件
操作

compare to dygraph quant.

上级 a95d4932
隐藏空白更改
内联 并排
Showing with 155 addition and 100 deletion
PaddleSlim/quant_low_level_api/quant.py
浏览文件 @ c7e09860
...@@ -158,7 +158,13 @@ def build_program(is_train, main_prog, startup_prog, args): ...@@ -158,7 +158,13 @@ def build_program(is_train, main_prog, startup_prog, args):
dtypes=["float32", "int64"], dtypes=["float32", "int64"],
use_double_buffer=True) use_double_buffer=True)
with fluid.unique_name.guard(): with fluid.unique_name.guard():
image, label = fluid.layers.read_file(py_reader) image = fluid.data(name="image", shape=[-1] + image_shape, dtype='float32')
label = fluid.data(name="label", shape=[-1, 1], dtype="int64")
# if is_train:
# image = fluid.data(name="image", shape=[-1] + image_shape, dtype='float32')
# label = fluid.data(name="label", shape=[-1, 1], dtype="int64")
# else:
# image, label = fluid.layers.read_file(py_reader)
out, avg_cost, acc_top1, acc_top5 = net_config(image, label, model, args) out, avg_cost, acc_top1, acc_top5 = net_config(image, label, model, args)
avg_cost.persistable = True avg_cost.persistable = True
acc_top1.persistable = True acc_top1.persistable = True
...@@ -171,11 +177,13 @@ def build_program(is_train, main_prog, startup_prog, args): ...@@ -171,11 +177,13 @@ def build_program(is_train, main_prog, startup_prog, args):
params["learning_strategy"]["batch_size"] = args.batch_size params["learning_strategy"]["batch_size"] = args.batch_size
params["learning_strategy"]["name"] = args.lr_strategy params["learning_strategy"]["name"] = args.lr_strategy
optimizer = optimizer_setting(params) # optimizer = optimizer_setting(params)
optimizer = fluid.optimizer.SGD(
learning_rate=0.1)
optimizer.minimize(avg_cost) optimizer.minimize(avg_cost)
global_lr = optimizer._global_learning_rate() global_lr = optimizer._global_learning_rate()
if is_train: if is_train:
return image, out, py_reader, avg_cost, acc_top1, acc_top5, global_lr return image, label, out, py_reader, avg_cost, acc_top1, acc_top5, global_lr
else: else:
return image, out, py_reader, avg_cost, acc_top1, acc_top5 return image, out, py_reader, avg_cost, acc_top1, acc_top5
...@@ -195,12 +203,12 @@ def train(args): ...@@ -195,12 +203,12 @@ def train(args):
train_prog = fluid.Program() train_prog = fluid.Program()
test_prog = fluid.Program() test_prog = fluid.Program()
_, _, train_py_reader, train_cost, train_acc1, train_acc5, global_lr = build_program( image, label, out, train_py_reader, train_cost, train_acc1, train_acc5, global_lr = build_program(
is_train=True, is_train=True,
main_prog=train_prog, main_prog=train_prog,
startup_prog=startup_prog, startup_prog=startup_prog,
args=args) args=args)
image, out, test_py_reader, test_cost, test_acc1, test_acc5 = build_program( _, _, test_py_reader, test_cost, test_acc1, test_acc5 = build_program(
is_train=False, is_train=False,
main_prog=test_prog, main_prog=test_prog,
startup_prog=startup_prog, startup_prog=startup_prog,
...@@ -233,12 +241,13 @@ def train(args): ...@@ -233,12 +241,13 @@ def train(args):
test_batch_size = 1 if activation_quant_type == 'abs_max' else 8 test_batch_size = 1 if activation_quant_type == 'abs_max' else 8
train_reader = paddle.batch( train_reader = paddle.batch(
reader.train(data_dir=data_dir), batch_size=train_batch_size, drop_last=True) reader.train(data_dir=data_dir), batch_size=train_batch_size, drop_last=True)
test_reader = paddle.batch(reader.val(data_dir=data_dir), batch_size=test_batch_size) test_reader = paddle.batch(reader.val(data_dir=data_dir), batch_size=128)
train_py_reader.decorate_paddle_reader(train_reader) # train_py_reader.decorate_paddle_reader(train_reader)
feeder_train = fluid.DataFeeder(
feed_list=[image, label], place=place)
test_py_reader.decorate_paddle_reader(test_reader) test_py_reader.decorate_paddle_reader(test_reader)
train_fetch_list = [train_cost.name, train_acc1.name, train_acc5.name, global_lr.name]
test_fetch_list = [test_cost.name, test_acc1.name, test_acc5.name] test_fetch_list = [test_cost.name, test_acc1.name, test_acc5.name]
# 1. Make some quantization transforms in the graph before training and testing. # 1. Make some quantization transforms in the graph before training and testing.
...@@ -261,77 +270,122 @@ def train(args): ...@@ -261,77 +270,122 @@ def train(args):
loss_name=train_cost.name, build_strategy=build_strategy) loss_name=train_cost.name, build_strategy=build_strategy)
test_prog = test_graph.to_program() test_prog = test_graph.to_program()
params = models.__dict__[args.model]().params params = models.__dict__[args.model]().params
fetch_var_name = 'conv1_weights'
train_fetch_list = [fetch_var_name, train_cost.name, train_acc1.name, train_acc5.name, global_lr.name]
for pass_id in range(params["num_epochs"]): for pass_id in range(params["num_epochs"]):
train_py_reader.start()
train_info = [[], [], []] train_info = [[], [], []]
test_info = [[], [], []] test_info = [[], [], []]
train_time = [] train_time = []
batch_id = 0 batch_id = 0
try:
while True: for batch_id, data in enumerate(test_reader()):
t1 = time.time() t1 = time.time()
loss, acc1, acc5, lr = exe.run(binary, fetch_list=train_fetch_list) # print(np.array(data)[0][0][0][0][:10])
t2 = time.time() x_data = np.array(
period = t2 - t1 [x[0].reshape(3, 224, 224) for x in data]).astype('float32')
loss = np.mean(np.array(loss)) # x_data = np.ones_like(np.array(x_data)) * batch_id
acc1 = np.mean(np.array(acc1)) # print(x_data[0][0][0][:10])
acc5 = np.mean(np.array(acc5)) y_data = np.array([x[1] for x in data]).astype('int64').reshape(-1, 1)
train_info[0].append(loss) w_v = np.array(fluid.global_scope().find_var('conv1_weights').get_tensor())
train_info[1].append(acc1) # print("weight check----------------", w_v[0][0][0][:10])
train_info[2].append(acc5) # img_data, loss, acc1, acc5, lr = exe.run(binary, feed=feeder_train.feed(data), fetch_list=train_fetch_list)
lr = np.mean(np.array(lr)) target_var, loss, acc1, acc5, lr = exe.run(binary, feed={image.name: x_data, label.name: y_data}, fetch_list=train_fetch_list)
train_time.append(period) # print("var check(name: {})----------------".format(fetch_var_name), np.array(target_var)[0][0][0][:10])
if batch_id % 10 == 0: t2 = time.time()
print("Pass {0}, trainbatch {1}, loss {2}, \ period = t2 - t1
acc1 {3}, acc5 {4}, lr {5}, time {6}" loss = np.mean(np.array(loss))
.format(pass_id, batch_id, loss, acc1, acc5, "%.6f" % acc1 = np.mean(np.array(acc1))
lr, "%2.2f sec" % period)) acc5 = np.mean(np.array(acc5))
sys.stdout.flush() train_info[0].append(loss)
batch_id += 1 train_info[1].append(acc1)
except fluid.core.EOFException: train_info[2].append(acc5)
train_py_reader.reset() lr = np.mean(np.array(lr))
train_time.append(period)
if batch_id % 1 == 0:
print("Pass {0}, trainbatch {1}, loss {2}, \
acc1 {3}, acc5 {4}, lr {5}, time {6}"
.format(pass_id, batch_id, loss, acc1, acc5, "%.6f" %
lr, "%2.2f sec" % period))
sys.stdout.flush()
batch_id += 1
train_loss = np.array(train_info[0]).mean() train_loss = np.array(train_info[0]).mean()
train_acc1 = np.array(train_info[1]).mean() train_acc1 = np.array(train_info[1]).mean()
train_acc5 = np.array(train_info[2]).mean() train_acc5 = np.array(train_info[2]).mean()
test_py_reader.start()
# for pass_id in range(params["num_epochs"]):
test_batch_id = 0
try: # train_py_reader.start()
while True:
t1 = time.time() # train_info = [[], [], []]
loss, acc1, acc5 = exe.run(program=test_prog, # test_info = [[], [], []]
fetch_list=test_fetch_list) # train_time = []
t2 = time.time() # batch_id = 0
period = t2 - t1 # try:
loss = np.mean(loss) # while True:
acc1 = np.mean(acc1) # t1 = time.time()
acc5 = np.mean(acc5) # img_data, loss, acc1, acc5, lr = exe.run(binary, fetch_list=train_fetch_list)
test_info[0].append(loss) # print(np.array(img_data)[0][0][0][:10])
test_info[1].append(acc1) # t2 = time.time()
test_info[2].append(acc5) # period = t2 - t1
if test_batch_id % 10 == 0: # loss = np.mean(np.array(loss))
print("Pass {0},testbatch {1},loss {2}, \ # acc1 = np.mean(np.array(acc1))
acc1 {3},acc5 {4},time {5}" # acc5 = np.mean(np.array(acc5))
.format(pass_id, test_batch_id, loss, acc1, acc5, # train_info[0].append(loss)
"%2.2f sec" % period)) # train_info[1].append(acc1)
sys.stdout.flush() # train_info[2].append(acc5)
test_batch_id += 1 # lr = np.mean(np.array(lr))
except fluid.core.EOFException: # train_time.append(period)
test_py_reader.reset() # if batch_id % 10 == 0:
# print("Pass {0}, trainbatch {1}, loss {2}, \
test_loss = np.array(test_info[0]).mean() # acc1 {3}, acc5 {4}, lr {5}, time {6}"
test_acc1 = np.array(test_info[1]).mean() # .format(pass_id, batch_id, loss, acc1, acc5, "%.6f" %
test_acc5 = np.array(test_info[2]).mean() # lr, "%2.2f sec" % period))
# sys.stdout.flush()
print("End pass {0}, train_loss {1}, train_acc1 {2}, train_acc5 {3}, " # batch_id += 1
"test_loss {4}, test_acc1 {5}, test_acc5 {6}".format( # except fluid.core.EOFException:
pass_id, train_loss, train_acc1, train_acc5, test_loss, # train_py_reader.reset()
test_acc1, test_acc5))
sys.stdout.flush() # train_loss = np.array(train_info[0]).mean()
# train_acc1 = np.array(train_info[1]).mean()
# train_acc5 = np.array(train_info[2]).mean()
# test_py_reader.start()
# test_batch_id = 0
# try:
# while True:
# t1 = time.time()
# loss, acc1, acc5 = exe.run(program=test_prog,
# fetch_list=test_fetch_list)
# t2 = time.time()
# period = t2 - t1
# loss = np.mean(loss)
# acc1 = np.mean(acc1)
# acc5 = np.mean(acc5)
# test_info[0].append(loss)
# test_info[1].append(acc1)
# test_info[2].append(acc5)
# if test_batch_id % 10 == 0:
# print("Pass {0},testbatch {1},loss {2}, \
# acc1 {3},acc5 {4},time {5}"
# .format(pass_id, test_batch_id, loss, acc1, acc5,
# "%2.2f sec" % period))
# sys.stdout.flush()
# test_batch_id += 1
# except fluid.core.EOFException:
# test_py_reader.reset()
# test_loss = np.array(test_info[0]).mean()
# test_acc1 = np.array(test_info[1]).mean()
# test_acc5 = np.array(test_info[2]).mean()
# print("End pass {0}, train_loss {1}, train_acc1 {2}, train_acc5 {3}, "
# "test_loss {4}, test_acc1 {5}, test_acc5 {6}".format(
# pass_id, train_loss, train_acc1, train_acc5, test_loss,
# test_acc1, test_acc5))
# sys.stdout.flush()
save_checkpoint_path = os.path.join(model_save_dir, model_name, str(pass_id)) save_checkpoint_path = os.path.join(model_save_dir, model_name, str(pass_id))
if not os.path.isdir(save_checkpoint_path): if not os.path.isdir(save_checkpoint_path):
...@@ -347,39 +401,39 @@ def train(args): ...@@ -347,39 +401,39 @@ def train(args):
# 2. Freeze the graph after training by adjusting the quantize # 2. Freeze the graph after training by adjusting the quantize
# operators' order for the inference. # operators' order for the inference.
freeze_pass = QuantizationFreezePass( # freeze_pass = QuantizationFreezePass(
scope=fluid.global_scope(), # scope=fluid.global_scope(),
place=place, # place=place,
weight_quantize_type=weight_quant_type) # weight_quantize_type=weight_quant_type)
freeze_pass.apply(test_graph) # freeze_pass.apply(test_graph)
server_program = test_graph.to_program() # server_program = test_graph.to_program()
fluid.io.save_inference_model( # fluid.io.save_inference_model(
dirname=float_path, # dirname=float_path,
feeded_var_names=[image.name], # feeded_var_names=[image.name],
target_vars=[out], executor=exe, # target_vars=[out], executor=exe,
main_program=server_program) # main_program=server_program)
# 3. Convert the weights into int8_t type. # 3. Convert the weights into int8_t type.
# (This step is optional.) # (This step is optional.)
convert_int8_pass = ConvertToInt8Pass(scope=fluid.global_scope(), place=place) # convert_int8_pass = ConvertToInt8Pass(scope=fluid.global_scope(), place=place)
convert_int8_pass.apply(test_graph) # convert_int8_pass.apply(test_graph)
server_int8_program = test_graph.to_program() # server_int8_program = test_graph.to_program()
fluid.io.save_inference_model( # fluid.io.save_inference_model(
dirname=int8_path, # dirname=int8_path,
feeded_var_names=[image.name], # feeded_var_names=[image.name],
target_vars=[out], executor=exe, # target_vars=[out], executor=exe,
main_program=server_int8_program) # main_program=server_int8_program)
# 4. Convert the freezed graph for paddle-mobile execution. # 4. Convert the freezed graph for paddle-mobile execution.
# (This step is optional.) # (This step is optional.)
mobile_pass = TransformForMobilePass() # mobile_pass = TransformForMobilePass()
mobile_pass.apply(test_graph) # mobile_pass.apply(test_graph)
mobile_program = test_graph.to_program() # mobile_program = test_graph.to_program()
fluid.io.save_inference_model( fluid.io.save_inference_model(
dirname=mobile_path, dirname="quant_infer",
feeded_var_names=[image.name], feeded_var_names=[image.name],
target_vars=[out], executor=exe, target_vars=[out], executor=exe,
main_program=mobile_program) main_program=test_graph.to_program())
def main(): def main():
args = parser.parse_args() args = parser.parse_args()
......
PaddleSlim/quant_low_level_api/run_quant.sh
浏览文件 @ c7e09860
...@@ -5,7 +5,7 @@ root_url="https://paddle-inference-dist.bj.bcebos.com/int8/pretrain" ...@@ -5,7 +5,7 @@ root_url="https://paddle-inference-dist.bj.bcebos.com/int8/pretrain"
MobileNetV1="MobileNetV1_pretrained.zip" MobileNetV1="MobileNetV1_pretrained.zip"
ResNet50="ResNet50_pretrained.zip" ResNet50="ResNet50_pretrained.zip"
GoogleNet="GoogleNet_pretrained.tar" GoogleNet="GoogleNet_pretrained.tar"
data_dir='Your image dataset path, e.g. ILSVRC2012' data_dir='/work/datasets/ILSVRC2012/'
pretrain_dir='../pretrain' pretrain_dir='../pretrain'
if [ ! -d ${pretrain_dir} ]; then if [ ! -d ${pretrain_dir} ]; then
...@@ -32,15 +32,16 @@ fi ...@@ -32,15 +32,16 @@ fi
cd - cd -
export CUDA_VISIBLE_DEVICES=0,1,2,3 export CUDA_VISIBLE_DEVICES=1
export FLAGS_cudnn_deterministic=1
#MobileNet v1: #MobileNet v1:
python quant.py \ python -u quant.py \
--model=MobileNet \ --model=MobileNet \
--pretrained_fp32_model=${pretrain_dir}/MobileNetV1_pretrained \ --pretrained_fp32_model=${pretrain_dir}/MobileNetV1_pretrained \
--use_gpu=True \ --use_gpu=True \
--data_dir=${data_dir} \ --data_dir=${data_dir} \
--batch_size=256 \ --batch_size=128 \
--total_images=1281167 \ --total_images=1281167 \
--class_dim=1000 \ --class_dim=1000 \
--image_shape=3,224,224 \ --image_shape=3,224,224 \
...@@ -48,7 +49,7 @@ python quant.py \ ...@@ -48,7 +49,7 @@ python quant.py \
--lr_strategy=piecewise_decay \ --lr_strategy=piecewise_decay \
--num_epochs=20 \ --num_epochs=20 \
--lr=0.0001 \ --lr=0.0001 \
--act_quant_type=abs_max \ --act_quant_type=moving_average_abs_max \
--wt_quant_type=abs_max --wt_quant_type=abs_max
......
PaddleSlim/reader.py
浏览文件 @ c7e09860
...@@ -11,7 +11,7 @@ np.random.seed(0) ...@@ -11,7 +11,7 @@ np.random.seed(0)
DATA_DIM = 224 DATA_DIM = 224
THREAD = 16 THREAD = 1
BUF_SIZE = 10240 BUF_SIZE = 10240
DATA_DIR = 'data/ILSVRC2012' DATA_DIR = 'data/ILSVRC2012'
...@@ -175,7 +175,7 @@ def train(data_dir=DATA_DIR): ...@@ -175,7 +175,7 @@ def train(data_dir=DATA_DIR):
return _reader_creator( return _reader_creator(
file_list, file_list,
'train', 'train',
shuffle=True, shuffle=False,
color_jitter=False, color_jitter=False,
rotate=False, rotate=False,
data_dir=data_dir) data_dir=data_dir)
......
Markdown is supported
0% 或 .
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 或