Skip to content

P
Paddle

PaddlePaddle / Paddle
11 个月 前同步成功

通知 2290
Star 20931
Fork 5422
P
Paddle

体验新版 GitCode，发现更多精彩内容 >>

未验证 提交 da051350 编写于 作者: C caozhou 提交者： GitHub
浏览文件
操作

[Auto Parallel] Add cost interface (#47043) 

* add cost interface

* update inferface and add unittest

* update unittest

* update inferface
上级 30dae6db
隐藏空白更改
内联 并排
Showing with 375 addition and 70 deletion
python/paddle/distributed/auto_parallel/cost/estimate_cost.py
浏览文件 @ da051350
......@@ -44,6 +44,8 @@ class CostEstimator:
) # {`op_id`: {"reshard": [], "dist_op": [], "local_cost": local_cost}}}
self._bubble_time_mapping = {}
self._ordered_ops = []
self.max_memories = {}
self.max_memory = None
@property
def loop_count(self):
......@@ -122,7 +124,7 @@ class CostEstimator:
for i in range(loop_count):
for op in ops:
self._detailed_cost[op.desc.id()] = OrderedDict()
# if in the while sub block, the detail of cost is the last cost
# If in the while sub block, the detail of cost is the last cost
detail = self._detailed_cost[op.desc.id()]
detail["reshard_cost"] = OrderedDict() #
detail["dist_op_cost"] = []
......@@ -146,15 +148,15 @@ class CostEstimator:
var = get_var_with_recursion(var_name, block, self.program)
reshard_cost = resharder.get_cost(op, var, self.cluster)
# calc reshard cost
# Calc reshard cost
if reshard_cost is not None:
detail["reshard_cost"][var_name] = reshard_cost
comm_costs = reshard_cost[0]
local_comp_cost = reshard_cost[1]
for comm_cost in comm_costs:
# time is cumulative in global cost and local cost, but memory and flops just are cumulative in global cost.
# comm sync
# Time is cumulative in global cost and local cost, but memory and flops just are cumulative in global cost.
# Comm sync
for item in comm_cost:
group_ranks, cost = item
max_time = None
......@@ -182,7 +184,7 @@ class CostEstimator:
for comp_cost in local_comp_cost[rank]:
self.local_cost(rank).time += comp_cost.time
# calc dist op cost
# Calc dist op cost
dist_op = dist_context.get_dist_op_for_program(op)
op_dist_attr = dist_op.dist_attr
processes = op_dist_attr.process_mesh.processes
......@@ -200,7 +202,7 @@ class CostEstimator:
continue
for item in dist_op_cost:
if isinstance(item, list):
# comm sync
# Comm sync
for comm_op_cost in item:
max_time = None
cost_time = {}
......@@ -221,9 +223,9 @@ class CostEstimator:
self._bubble_time_mapping[rank] += (
max_time - cost_time[rank])
elif isinstance(item, dict):
# op just one
# Op just one
for rank in processes:
# dp+pp+mp
# DP+PP+MP
if rank not in item:
continue
self.local_cost(rank).time += item[rank].time
......@@ -266,7 +268,7 @@ class CostEstimator:
return result
memories = {}
max_memories = {}
self.max_memories = {}
var_info = {
} # var_name: [[process_mesh, dims_mapping], [id]], [[process_mesh, dims_mapping], [id]]}
......@@ -276,6 +278,10 @@ class CostEstimator:
self._ordered_ops.sort(key=lambda x: x[0])
for op_id, op in self._ordered_ops:
if op.type in [
"create_py_reader", "create_double_buffer_reader", "read"
]:
continue
dist_op = dist_context.get_dist_op_for_program(op)
process_mesh = dist_op.dist_attr.process_mesh
for var_name in op.input_arg_names:
......@@ -287,7 +293,7 @@ class CostEstimator:
input_dims_mapping)
if key not in var_info[var_name]:
var_info[var_name][key] = {}
# it is even partition now
# It is even partition now
if "memory" not in var_info[var_name][key]:
var = dist_op.get_serial_input(var_name)
global_sizes = var.shape
......@@ -325,6 +331,10 @@ class CostEstimator:
has_used_vars = set()
for op_id, op in self._ordered_ops:
if op.type in [
"create_py_reader", "create_double_buffer_reader", "read"
]:
continue
can_free_memories = {}
can_free_vars = set()
dist_op = dist_context.get_dist_op_for_program(op)
......@@ -336,14 +346,14 @@ class CostEstimator:
input_dims_mapping)
has_used_var = var_name + key
var = dist_op.get_serial_input(var_name)
# not used
# Not used
if var_name + key not in has_used_vars:
has_used_vars.add(has_used_var)
for process in process_mesh.processes:
if process not in memories:
memories[process] = 0
memories[process] += var_info[var_name][key]["memory"]
# used
# Used
else:
if op_id == var_info[var_name][key]["position"][-1]:
if has_used_var not in can_free_vars:
......@@ -362,14 +372,14 @@ class CostEstimator:
output_dims_mapping)
has_used_var = var_name + key
var = dist_op.get_serial_output(var_name)
# not used
# Not used
if var_name + key not in has_used_vars:
has_used_vars.add(has_used_var)
for process in process_mesh.processes:
if process not in memories:
memories[process] = 0
memories[process] += var_info[var_name][key]["memory"]
# used
# Used
else:
if op_id == var_info[var_name][key]["position"][-1]:
if has_used_var not in can_free_vars:
......@@ -381,21 +391,22 @@ class CostEstimator:
can_free_memories[process] += var_info[
var_name][key]["memory"]
# calc peak memory
# Calc peak memory
for process in memories:
if process not in max_memories:
max_memories[process] = memories[process]
if process not in self.max_memories:
self.max_memories[process] = memories[process]
else:
if memories[process] > max_memories[process]:
max_memories[process] = memories[process]
if memories[process] > self.max_memories[process]:
self.max_memories[process] = memories[process]
# free memory
# Free memory
for process in can_free_memories:
if process in memories:
memories[process] -= can_free_memories[process]
# Calculate the max memory in all ranks
max_memory = max(max_memories.values())
max_memory = max(self.max_memories.values())
self.max_memory = max_memory
return max_memory
......@@ -409,3 +420,143 @@ class CostEstimator:
self._estimate_core(dist_context, resharder, block)
return self.global_cost
def _print_tag(self, max_len, length):
tag = "+" + "-" * max_len
for i in range(length):
print(tag, end="")
if i == length - 1:
print("+")
def _print_vals(self, vals, max_len):
for idx, val in enumerate(vals):
s = "|" + str(val).center(max_len)
print(s, end="")
if idx == len(vals) - 1:
print("|")
def _pretty_print_memory_cost(self):
"""Print memory of every rank prettily."""
if not self.max_memories or not self.max_memory:
raise ValueError("Please calculate memory cost before print.")
# Padding automatically
max_len = 0
header = ["Rank", "Memory(MiB)"]
memories = [
int(item // 1e6) for item in list(self.max_memories.values())
]
for memory in (memories + header):
if len(str(memory)) > max_len:
max_len = len(str(memory))
max_len += 4 # for pretty print of center
# Print tag
self._print_tag(max_len, len(header))
# Print header
self._print_vals(header, max_len)
# Print tag
self._print_tag(max_len, len(header))
# Print rank and its memory
for i in range(len(self.max_memories)):
memory = memories[i]
vals = [i, memory]
self._print_vals(vals, max_len)
self._print_tag(max_len, len(header))
def _pretty_print_global(self):
"""Print global execution time and max memory prettily."""
if not self.max_memories or not self.max_memory:
raise ValueError("Please calculate cost before print.")
# Padding automatically
max_len = 0
header = ["Execution Time(ms)", "Max Memory(MiB)"]
vals = [round(self.global_cost.time, 3), int(self.max_memory // 1e6)]
for memory in (vals + header):
if len(str(memory)) > max_len:
max_len = len(str(memory))
max_len += 4 # for pretty print of center
# Print tag
self._print_tag(max_len, len(header))
# Print header
self._print_vals(header, max_len)
# Print tag
self._print_tag(max_len, len(header))
# Print exec time and max memory
self._print_vals(vals, max_len)
# Print tag
self._print_tag(max_len, len(header))
def pretty_print_cost(self):
"""Print cost prettily."""
print("The global execution time and max memory are as follows:")
self._pretty_print_global()
print("The memory of every rank is as follows:")
self._pretty_print_memory_cost()
def get_cost_from_engine(engine, mode):
from ..utils import to_list
# Construct cost estimator by original main program
serial_main_prog = engine._serial_main_progs[mode].clone(
) if mode in engine._serial_main_progs else engine._orig_main_prog.clone()
serial_startup_prog = engine._serial_startup_progs[mode].clone(
) if mode in engine._serial_startup_progs else engine._orig_startup_prog.clone(
)
losses = to_list(
engine._loss) if (not isinstance(engine._loss, paddle.nn.Layer)
and not callable(engine._loss)) else engine._losses
if mode in engine._dist_contexts:
dist_context = engine._dist_contexts[mode]
completer = engine._planners[mode].completer
else:
from ..completion import Completer
from ..dist_context import DistributedContext
dist_context = DistributedContext(serial_main_prog, serial_startup_prog,
engine._optimizer, losses, {},
{"loss": losses}, engine._cluster,
engine._strategy)
completer = Completer(dist_context)
completer.complete_forward_annotation()
dist_context.block_state.parse_forward_blocks(
dist_context.serial_main_program)
if mode == "eval" or mode == "predict":
cost_estimator = CostEstimator(serial_main_prog, engine._cluster)
elif mode == "train":
from ..parallelizer_v2 import Parallelizer
# Get serial main program with backward
serial_optimizer = engine._optimizer
parallelizer = Parallelizer(mode, completer, dist_context)
# Generate backward
loss_name = dist_context.serial_loss.name
serial_loss = serial_main_prog.global_block()._var_recursive(loss_name)
params_grads = parallelizer._generate_backward(serial_main_prog,
serial_startup_prog,
serial_loss)
# Generate optimizer
optimizer_ops = parallelizer._generate_optimizer(
serial_main_prog, serial_startup_prog, serial_optimizer,
params_grads)
cost_estimator = CostEstimator(serial_main_prog, engine._cluster)
# Estimate global_cost and max memory
global_cost = cost_estimator.estimate(dist_context)
max_memory = cost_estimator._estimate_max_memory_by_dist_op(dist_context)
# Print the cost
cost_estimator.pretty_print_cost()
return global_cost, max_memory
python/paddle/distributed/auto_parallel/engine.py
浏览文件 @ da051350
......@@ -48,6 +48,8 @@ from .dist_context import DistributedContext, get_default_distributed_context
from .strategy import Strategy
from .interface import CollectionNames, get_collection
from ..utils.log_utils import get_logger
from .utils import initialize_pg_in_full_mode
from .cost.estimate_cost import get_cost_from_engine
class Engine:
......@@ -127,12 +129,6 @@ class Engine:
"'model must be sub classes of `paddle.nn.Layer` or any callable function."
)
self._model = model
# if loss and not isinstance(loss,
# paddle.nn.Layer) and not callable(loss):
# raise TypeError(
# "'loss' must be sub classes of `paddle.nn.Layer` or any callable function."
# )
self._loss = loss
if optimizer and not isinstance(
......@@ -201,6 +197,7 @@ class Engine:
self._planned_mode = None
self._dygraph_mode = False
self._tuning = self._strategy.tuning
self._losses = None
self.history = None
......@@ -487,6 +484,7 @@ class Engine:
outputs = self.program_helper.output_vars
labels = self.program_helper.label_vars
losses = self.program_helper.loss_vars
self._losses = losses
metrics = self.program_helper.metric_vars
self._inputs = inputs
......@@ -512,6 +510,7 @@ class Engine:
outputs = to_list(self._model(*inputs))
if mode != "predict" and self._loss:
losses = to_list(self._loss(*(outputs + labels)))
self._losses = losses
if mode != "predict" and (outputs or labels):
for metric in self._metrics:
......@@ -519,6 +518,7 @@ class Engine:
to_list(metric.compute(*(outputs + labels))))
else:
losses = to_list(self._loss)
self.losses = losses
default_ctx = get_default_distributed_context()
if not default_ctx.has_annotation:
......@@ -648,11 +648,13 @@ class Engine:
# instantiate communication by process_mapping.
all_process_groups = get_all_process_groups()
# NOTE: add the comm init control in the future for auto search
for process_group in all_process_groups:
if self._cur_rank not in process_group.ranks:
continue
process_group.instantiate()
if self._strategy.auto_mode == "full":
initialize_pg_in_full_mode(all_process_groups, cur_rank)
else:
for process_group in all_process_groups:
if self._cur_rank not in process_group.ranks:
continue
process_group.instantiate()
place = _get_device()
if isinstance(place, fluid.CUDAPlace):
......@@ -1022,16 +1024,9 @@ class Engine:
test_dataloader = self._prepare_dataloader_from_generator(
dataset=test_data,
# feed_list=feed_list,
capacity=70,
# use_double_buffer=use_double_buffer,
iterable=False,
# return_list=return_list,
# use_multiprocess=use_multiprocess,
# drop_last=drop_last,
# places=places,
batch_size=batch_size,
# epochs=epochs,
steps_per_epoch=steps,
collate_fn=collate_fn)
......@@ -1100,21 +1095,19 @@ class Engine:
steps_per_epoch=steps_per_epoch)
return dataloader
def dataloader_from_generator(
self,
dataset,
capacity=70,
use_double_buffer=True,
iterable=True,
# return_list=False,
use_multiprocess=False,
drop_last=True,
batch_size=1,
epochs=1,
steps_per_epoch=None,
collate_fn=None,
sample_split=1,
mode=None):
def dataloader_from_generator(self,
dataset,
capacity=70,
use_double_buffer=True,
iterable=True,
use_multiprocess=False,
drop_last=True,
batch_size=1,
epochs=1,
steps_per_epoch=None,
collate_fn=None,
sample_split=1,
mode=None):
if mode is not None:
self.to_mode(mode)
self._inputs_spec, self._labels_spec = self._prepare_data_spec(
......@@ -1127,14 +1120,12 @@ class Engine:
self._switch_mode(self._mode)
dataloader = self._prepare_dataloader_from_generator(
dataset=dataset,
# feed_list=feed_list,
capacity=capacity,
use_double_buffer=use_double_buffer,
iterable=iterable,
return_list=False,
use_multiprocess=use_multiprocess,
drop_last=drop_last,
# places=places,
batch_size=batch_size,
epochs=epochs,
steps_per_epoch=steps_per_epoch,
......@@ -1187,20 +1178,7 @@ class Engine:
assert self._inputs_spec and self._labels_spec, \
"Please call the dataloader(...) before calling prepare(...)"
def run(
self,
data=None,
# program=None,
feed=None,
fetch_list=None,
# feed_var_name='feed',
# fetch_var_name='fetch',
# scope=None,
# return_numpy=True,
# use_program_cache=False,
# return_merged=True,
# use_prune=False,
mode=None):
def run(self, data=None, feed=None, fetch_list=None, mode=None):
if mode is not None:
self.to_mode(mode)
feed_dict = self._prepare_feed(data, feed, self._mode)
......@@ -1571,6 +1549,54 @@ class Engine:
path, load_optimizer)
return self._state_dict, self._dist_attr
def cost(self, inputs_spec=None, labels_spec=None, mode="train"):
"""
Get and Print cost, including memory of every rank,
max memory among all ranks, and the global cost of one step based on
communication cost(computation cost is 0 by default).
In the future, the flops information of every rank and global cost including
computation cost will be added.
Args:
inputs_spec(InputSpec): The specification of inputs. Default: None.
labels_spec(InputSpec): The specification of labels. Default: None.
mode (str): The engine mode must be in ["train", "predict", "eval"]. Default: "train".
Returns:
Return the global execution time (ms) and max memory (B).
"""
# Check parallel mode
if self._strategy.auto_mode == "full":
print(
"The cost will be calcudated in the search process when the auto mode is full."
)
return
# Check mode
accepted_modes = ["train", "predict", "eval"]
if mode not in accepted_modes:
raise ValueError("The mode {} is not in accepted modes {}".format(
mode, accepted_modes))
self.to_mode(mode)
if inputs_spec is not None:
self._inputs_spec, self._labels_spec = inputs_spec, labels_spec
self._inputs, self._labels = self._prepare_data_tensor(
self._inputs_spec, self._labels_spec)
self._build(mode)
self._plan(mode)
else:
if _non_static_mode() or self._dygraph_mode:
raise ValueError(
"Please call `engine._prepare_program('mode')` firstly when in the static graph mode."
)
# Estimate the exec cost and max memory
global_cost, max_memory = get_cost_from_engine(self, mode)
return global_cost.time, max_memory
@property
def main_program(self):
return self._dist_main_progs[self._mode][self._cur_rank]
......
python/paddle/distributed/auto_parallel/utils.py
浏览文件 @ da051350
......@@ -1602,3 +1602,65 @@ def get_lr(optimizer):
"'optimizer' must be object of class `paddle.optimizer.Optimizer`" \
" or `paddle.fluid.optimizer.Optimizer`, but got {}.".format(type(optimizer))
)
def initialize_pg_in_full_mode(all_process_groups, cur_rank):
import socket
from ..collective import _get_global_env
has_recv_by_socket = []
# This is a magic number
magic_num = 500
genv = _get_global_env()
cur_rank_ip, cur_rank_port = genv.current_endpoint.split(":")
cur_rank_recv_port = int(cur_rank_port) + magic_num
server_socket = None
# Large enough for recv rank
buff_size = 1024
server_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
server_socket.bind((cur_rank_ip, cur_rank_recv_port))
# The 10 is an empirical value
server_socket.listen(10)
client_sockets = {}
for process_group in all_process_groups:
if cur_rank not in process_group.ranks:
continue
if len(process_group.ranks) == 2:
index = process_group.ranks.index(cur_rank)
is_send = True if index == 0 else False
if is_send:
recv_rank = process_group.ranks[1]
recv_rank_ip, recv_rank_port = genv.trainer_endpoints[
recv_rank].split(":")
connect_port = int(recv_rank_port) + magic_num
client_socket = socket.socket(socket.AF_INET,
socket.SOCK_STREAM)
client_socket.connect((recv_rank_ip, connect_port))
client_socket.send(str(cur_rank).encode('utf-8'))
rank = client_socket.recv(buff_size).decode('utf-8')
rank = int(rank)
if rank != recv_rank:
raise ValueError(
"Please check comm pair, the recv rank should be {} but got {}."
.format(recv_rank, rank))
else:
print("It is able to instantiate {} as sender now.".format(
process_group.ranks))
client_socket.close()
else:
send_rank = process_group.ranks[0]
while True:
if send_rank not in has_recv_by_socket:
client_socket, recv_addr = server_socket.accept()
rank = int(client_socket.recv(buff_size).decode())
client_sockets[rank] = client_socket
has_recv_by_socket.append(rank)
else:
client_sockets[send_rank].send(
str(cur_rank).encode("utf-8"))
client_sockets[send_rank].close()
print("It is able to instantiate {} as recver now.".
format(process_group.ranks))
break
process_group.instantiate()
server_socket.close()
python/paddle/fluid/tests/unittests/auto_parallel/engine_api.py
浏览文件 @ da051350
......@@ -141,6 +141,7 @@ def train_high_level(fetch):
# train
train_dataset = MyDataset(batch_num * batch_size)
eval_dataset1 = MyDataset(5 * batch_size)
history = engine.fit(train_data=train_dataset,
epochs=2,
batch_size=batch_size,
......@@ -354,9 +355,74 @@ def train_non_builtin_data_vars():
) # call DataLoader.reset() after catching EOFException
def get_cost():
main_program = static.default_main_program()
startup_program = static.default_startup_program()
with static.program_guard(main_program,
startup_program), utils.unique_name.guard():
input = static.data(name="input",
shape=[batch_size, image_size],
dtype='float32')
label = static.data(name="label", shape=[batch_size, 1], dtype='int64')
loader = paddle.io.DataLoader.from_generator(feed_list=[input, label],
capacity=4 * batch_size,
iterable=False)
places = static.cuda_places()
loader.set_batch_generator(batch_generator_creator(), places=places)
mlp = MLPLayer(hidden_size=hidden_size,
intermediate_size=4 * hidden_size,
dropout_ratio=0.1,
initializer_range=0.02)
loss = paddle.nn.CrossEntropyLoss()
optimizer = paddle.optimizer.Adam(learning_rate=0.00001,
beta1=0.9,
beta2=0.999,
epsilon=1e-08,
grad_clip=None)
metric = paddle.metric.Accuracy()
predict = mlp(input)
loss_var = loss(predict, label)
strategy = auto.Strategy()
strategy.auto_mode = "semi"
engine = auto.Engine(loss=loss_var,
optimizer=optimizer,
metrics=metric,
strategy=strategy)
engine.cost()
def get_cost_by_spec():
mlp = MLPLayer(hidden_size=hidden_size,
intermediate_size=4 * hidden_size,
dropout_ratio=0.1,
initializer_range=0.02)
loss = paddle.nn.CrossEntropyLoss()
optimizer = paddle.optimizer.Adam(learning_rate=0.00001,
beta1=0.9,
beta2=0.999,
epsilon=1e-08,
grad_clip=None)
metric = paddle.metric.Accuracy()
strategy = auto.Strategy()
strategy.auto_mode = "semi"
engine = auto.Engine(mlp, loss, optimizer, metric, strategy=strategy)
input_spec = static.InputSpec([batch_size, image_size], 'float32', 'input')
label_spec = static.InputSpec([batch_size, 1], 'int64', 'label')
engine.cost(mode="eval", inputs_spec=[input_spec], labels_spec=[label_spec])
if __name__ == "__main__":
train_high_level(fetch=True)
train_high_level(fetch=False)
train_low_level()
train_builtin_data_vars()
train_non_builtin_data_vars()
get_cost()
get_cost_by_spec()
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