ColossalAI/colossalai/zero/sharded_optim/sharded_optim_v2.py

355 lines
17 KiB
Python
Raw Normal View History

2022-03-03 07:06:18 +00:00
from enum import Enum
from os import stat
from typing import Dict, Optional, Tuple
2022-03-03 07:06:18 +00:00
import torch
import torch.distributed as dist
import torch.nn as nn
from colossalai.amp.naive_amp.grad_scaler import DynamicGradScaler
2022-03-03 07:06:18 +00:00
from colossalai.context.parallel_mode import ParallelMode
from colossalai.core import global_context as gpc
from colossalai.logging import get_dist_logger
2022-03-03 07:06:18 +00:00
from colossalai.nn.optimizer import ColossalaiOptimizer
from colossalai.utils.memory_tracer.model_data_memtracer import \
GLOBAL_MODEL_DATA_TRACER
from colossalai.zero.shard_utils.tensor_utils import (colo_model_data_tensor_move_inline, colo_model_tensor_clone,
colo_tensor_mem_usage)
2022-03-03 07:06:18 +00:00
from colossalai.zero.sharded_model import ShardedModelV2
2022-03-25 06:54:39 +00:00
from colossalai.zero.sharded_model._utils import cast_tensor_to_fp32
from colossalai.zero.sharded_optim._utils import has_inf_or_nan
from colossalai.zero.sharded_param.tensorful_state import (StatefulTensor, TensorState)
from torch import Tensor
from torch.distributed import ProcessGroup
from torch.nn.parameter import Parameter
from torch.optim import Optimizer
2022-03-29 07:45:48 +00:00
2022-03-03 07:06:18 +00:00
class OptimState(Enum):
SCALED = 1
UNSCALED = 2
2022-03-03 07:55:27 +00:00
class ShardedOptimizerV2(ColossalaiOptimizer):
2022-04-01 06:50:56 +00:00
"""A wrapper for optimizer. ``ShardedOptimizerV2`` and ``ShardedModelV2`` implement Zero Redundancy Optimizer (ZeRO).
By default the ZeRO optimizer stage 3 offload Optimizer States on CPU.
We apply the Device-aware Operator Placement technique for OS placement from the following paper.
2022-04-01 06:50:56 +00:00
`PatrickStar: Parallel Training of Pre-trained Models via Chunk-based Memory Management`_
GPU margin space is the remaining space after removing peak non-model data from the overall GPU memory,
which is detected by a runtime memory tracer.
We place as many OS chunks in the margin space as possible.
2022-04-01 06:50:56 +00:00
The size of margin space can be controlled by ``gpu_margin_mem_ratio``.
If it is set as ``0.0``, it is the same as classical ZeRO optimizer.
2022-04-01 06:50:56 +00:00
Note:
You must use ``ShardedOptimizerV2`` with ``ShardedModelV2``.
Note:
Make sure you enable ``use_memory_tracer`` in ``ShardedModelV2``,
if you set ``gpu_margin_mem_ratio > 0``.
Args:
sharded_model (ShardedModelV2): A sharded model initialized by class ShardedModelV2. The optimizer will use the
shard strategy provided by sharded model to shard param fp32 tensors.
optimizer (Optimizer): An Optimizer instance.
cpu_offload (bool, optional): Is offloading the optimizer states to CPU.. Defaults to False.
gpu_margin_mem_ratio (float, optional): The ratio of GPU remaining memory (after the first forward-backward)
2022-04-01 06:50:56 +00:00
which will be used when using hybrid CPU optimizer.
Make sure `reuse_fp16_shard` is enabled in `ShardedModelV2`, if `gpu_margin_mem_ratio` > `0.0`.
Defaults to 0.0.
initial_scale (float, optional): Initial scale used by DynamicGradScaler. Defaults to 2**32.
min_scale (float, optional): Min scale used by DynamicGradScaler. Defaults to 1.
growth_factor (float, optional): growth_factor used by DynamicGradScaler. Defaults to 2.
backoff_factor (float, optional): backoff_factor used by DynamicGradScaler. Defaults to 0.5.
growth_interval (float, optional): growth_interval used by DynamicGradScaler. Defaults to 1000.
hysteresis (float, optional): hysteresis used by DynamicGradScaler. Defaults to 2.
keep_unsharded (bool, optional): if True, optimizer won't shard unsharded parameters.
In Zero-2, set keep_unsharded to False.
In Zero-3, set keep_unsharded to True.
max_scale (int, optional): max_scale used by DynamicGradScaler. Defaults to 2**32.
dp_process_group (Optional[ProcessGroup], optional): data paralle process group. Defaults to None.
mp_process_group (Optional[ProcessGroup], optional): model paralle process group. Defaults to None.
2022-04-01 06:50:56 +00:00
.. _PatrickStar\: Parallel Training of Pre-trained Models via Chunk-based Memory Management:
https://arxiv.org/abs/2108.05818
"""
2022-03-03 07:06:18 +00:00
def __init__(self,
2022-03-09 08:09:36 +00:00
sharded_model: ShardedModelV2,
optimizer: Optimizer,
2022-03-03 07:06:18 +00:00
cpu_offload: bool = False,
gpu_margin_mem_ratio: float = 0.0,
2022-03-03 07:06:18 +00:00
initial_scale: float = 2**32,
min_scale: float = 1,
growth_factor: float = 2,
backoff_factor: float = 0.5,
growth_interval: float = 1000,
hysteresis: float = 2,
max_scale: int = 2**32,
keep_unsharded: bool = False,
2022-03-03 07:06:18 +00:00
dp_process_group: Optional[ProcessGroup] = None,
mp_process_group: Optional[ProcessGroup] = None) -> None:
2022-03-09 08:09:36 +00:00
assert isinstance(sharded_model, ShardedModelV2), 'model must be wrapped with ShardedModel'
super().__init__(optimizer)
self.shard_strategy = sharded_model.shard_strategy
2022-03-09 08:09:36 +00:00
self.model: ShardedModelV2 = sharded_model
if cpu_offload and not sharded_model.cpu_offload:
raise RuntimeError(
f"ShardedOptimizerV2 using cpu_offload, but the sharded_model used to initialize it dose not use cpu_offload"
)
self.gpu_margin_mem_ratio: float = float(gpu_margin_mem_ratio)
assert 0.0 <= self.gpu_margin_mem_ratio <= 1.0, f'gpu_margin_mem_ratio must >=0.0 and <=1.0'
# Only move fp32 shards from CPU to GPU when user allows and inner optimizer is valid
# Inner optimizer must support optimizing hybrid (CPU and CUDA) tensors,
# and it must set `num_fp32_shards_per_param` correctly
self._should_move_fp32_shards_h2d: bool = cpu_offload and self.gpu_margin_mem_ratio > 0.0 and getattr(
optimizer, 'num_fp32_shards_per_param', 0) >= 2
2022-03-03 07:42:53 +00:00
self.device = torch.cuda.current_device() if not cpu_offload else torch.device('cpu')
2022-03-03 07:06:18 +00:00
self.optim_state: OptimState = OptimState.UNSCALED
self.dp_process_group = dp_process_group or gpc.get_group(ParallelMode.DATA)
self.mp_process_group = mp_process_group or gpc.get_group(ParallelMode.MODEL)
# Grad scaler
self.grad_scaler = DynamicGradScaler(initial_scale=initial_scale,
min_scale=min_scale,
growth_factor=growth_factor,
backoff_factor=backoff_factor,
growth_interval=growth_interval,
hysteresis=hysteresis,
max_scale=max_scale)
2022-03-09 08:09:36 +00:00
self._found_overflow: Tensor = torch.FloatTensor([0]).to(torch.cuda.current_device())
self._logger = get_dist_logger("ShardedOptimizerV2")
2022-03-03 07:42:53 +00:00
assert not (keep_unsharded and self._should_move_fp32_shards_h2d), \
"Keeping unsharded parameters can't be used with hybrid OS placement right now."
self.keep_unshard = keep_unsharded
2022-03-03 07:06:18 +00:00
# Store fp32 param shards
self._register_master_weight()
2022-03-03 07:06:18 +00:00
self._logger.debug(f"After init ShardedOptimizerV2 consumes {self.get_memory_usage()[0]/1e6} MB CUDA Memory!",
ranks=[0])
2022-03-29 07:45:48 +00:00
self._use_memory_tracer = self.model.use_memory_tracer
if self._use_memory_tracer:
GLOBAL_MODEL_DATA_TRACER.register_optimizer(self)
def get_memory_usage(self) -> Tuple[int, int]:
2022-04-01 06:50:56 +00:00
""" Get the memory usage of the optimizer. Including master_params (param fp32),
momentum (``self.state[p]['exp_avg']``) variance (``self.state[p]['exp_avg_sq']``)
Returns:
Tuple[int, int]: cuda/cpu memory usage in Byte.
"""
cuda_use = 0
cpu_use = 0
def update_mem_use(t):
nonlocal cuda_use
nonlocal cpu_use
t_cuda_use, t_cpu_use = colo_tensor_mem_usage(t)
cuda_use += t_cuda_use
cpu_use += t_cpu_use
for _, p_fp32 in self.master_params.items():
update_mem_use(p_fp32)
for group in self.optim.param_groups:
for p in group['params']:
state = self.optim.state[p]
for k, v in state.items():
update_mem_use(v)
return cuda_use, cpu_use
2022-03-03 07:06:18 +00:00
def step(self, *args, **kwargs):
self._prepare_grads()
self._maybe_move_fp32_shards()
2022-03-03 07:06:18 +00:00
# unscale grads if scaled
if self.optim_state == OptimState.SCALED:
self._unscale_grads()
found_inf = self._check_overflow()
self.grad_scaler.update(found_inf)
if found_inf:
self._logger.warning('found inf during ShardedOptimV2 step')
self._zero_grad(recover_data=True)
2022-03-03 07:06:18 +00:00
return
self._point_param_fp16_to_master_param()
2022-03-10 09:51:50 +00:00
self._logger.debug(
f"Before step ShardedOptimizerV2 consumes {self.get_memory_usage()[0]/1e6} MB CUDA Memory, {self.get_memory_usage()[1]/1e6} MB CUDA Memory!",
ranks=[0])
ret = self.optim.step(*args, **kwargs)
2022-03-10 09:51:50 +00:00
self._logger.debug(
f"After step ShardedOptimizerV2 consumes {self.get_memory_usage()[0]/1e6} MB CUDA Memory, {self.get_memory_usage()[1]/1e6} MB CUDA Memory!",
ranks=[0])
self._copy_master_param_to_param_fp16()
2022-03-03 07:06:18 +00:00
return ret
def backward(self, loss: Tensor) -> None:
loss = self.loss_scale * loss
self.optim_state = OptimState.SCALED
2022-03-09 08:09:36 +00:00
self.model.backward(loss)
2022-03-03 07:06:18 +00:00
def backward_by_grad(self, tensor: Tensor, grad: Tensor) -> None:
2022-03-09 08:09:36 +00:00
self.model.backward_by_grad(tensor, grad)
2022-03-03 07:06:18 +00:00
def clip_grad_norm(self, model: nn.Module, max_norm: float):
if self.optim_state == OptimState.SCALED:
self._unscale_grads()
return super().clip_grad_norm(model, max_norm)
@property
def loss_scale(self):
2022-03-09 08:09:36 +00:00
return self.grad_scaler.scale.item()
2022-03-03 07:06:18 +00:00
def _check_overflow(self):
# clear previous overflow record
self._found_overflow.fill_(0.0)
# check for overflow
for group in self.optim.param_groups:
2022-03-03 07:06:18 +00:00
for p in group['params']:
if has_inf_or_nan(p.grad):
self._found_overflow.fill_(1.0)
break
# all-reduce across dp group
dist.all_reduce(self._found_overflow, op=dist.ReduceOp.MAX, group=self.dp_process_group)
# all-reduce over model parallel group
dist.all_reduce(self._found_overflow, op=dist.ReduceOp.MAX, group=self.mp_process_group)
2022-03-03 07:50:30 +00:00
return self._found_overflow.item() > 0
2022-03-03 07:06:18 +00:00
def _unscale_grads(self):
assert self.optim_state == OptimState.SCALED
for group in self.optim.param_groups:
2022-03-03 07:06:18 +00:00
for p in group['params']:
if p.grad is not None:
p.grad.data.div_(self.loss_scale)
self.optim_state = OptimState.UNSCALED
2022-03-09 08:09:36 +00:00
def zero_grad(self, *args, **kwargs):
self._zero_grad()
def _zero_grad(self, recover_data: bool = False):
"""zero grad and maybe recover fp16 params
When `reuse_fp16_shard` is enabled,
p.colo_attr.sharded_data_tensor stores grad here.
We have to recover them from fp32 params.
Args:
recover_data (bool, optional): Whether to recover fp16 param from fp32 param. Defaults to False.
"""
2022-03-09 08:09:36 +00:00
# We must set grad to None
# Because grad here is sharded
# But next backward pass will create a full grad first
# Which leads to wrong accumulation
self.optim.zero_grad(set_to_none=True)
for group in self.optim.param_groups:
for p in group['params']:
# p.colo_attr.sharded_data_tensor stores grad now
# we have to recover fp16 param
reuse_fp16_shard = p.colo_attr.saved_grad.data_ptr() == p.colo_attr.sharded_data_tensor.data_ptr()
p.colo_attr.saved_grad.set_null()
if recover_data and reuse_fp16_shard:
p.colo_attr.sharded_data_tensor.reset_payload(
colo_model_tensor_clone(self.master_params[p].payload.half(), torch.cuda.current_device()))
def sync_grad(self):
pass
def _register_master_weight(self):
self.master_params: Dict[Parameter, StatefulTensor] = {}
for group in self.optim.param_groups:
for p in group['params']:
assert hasattr(p, 'colo_attr'), 'The parameter must be wrapped with ShardedParam'
is_param_sharded = p.colo_attr.sharded_data_tensor.is_sharded
if not is_param_sharded and not self.keep_unshard:
# Please use keep_unsharded to control whether shard unsharded paramters
# As we only store param shard, we shard it here
self.shard_strategy.shard([p.colo_attr.sharded_data_tensor], self.dp_process_group)
self.master_params[p] = StatefulTensor(
cast_tensor_to_fp32(p.colo_attr.sharded_data_tensor.payload).to(self.device))
if not is_param_sharded and not self.keep_unshard:
# In this branch, there's no need to shard param
# So we gather here
self.shard_strategy.gather([p.colo_attr.sharded_data_tensor], self.dp_process_group)
def _maybe_move_fp32_shards(self):
if self._should_move_fp32_shards_h2d:
self._should_move_fp32_shards_h2d = False
available_cuda_margin_mem = self.model.cuda_margin_space * self.gpu_margin_mem_ratio
fp32_shards_available_cuda_margin_mem = available_cuda_margin_mem / self.optim.num_fp32_shards_per_param
fp32_shards_used_cuda_margin_mem = 0
for group in self.optim.param_groups:
for p in group['params']:
shard_mem = self.master_params[p].payload.numel() * self.master_params[p].payload.element_size()
if fp32_shards_used_cuda_margin_mem + shard_mem < fp32_shards_available_cuda_margin_mem:
colo_model_data_tensor_move_inline(self.master_params[p], torch.cuda.current_device())
p.grad.data = p.grad.data.to(torch.cuda.current_device())
p.colo_attr.offload_grad = False
fp32_shards_used_cuda_margin_mem += shard_mem
def _prepare_grads(self):
for group in self.optim.param_groups:
for p in group['params']:
p.colo_attr.saved_grad.trans_state(TensorState.COMPUTE)
# FIXME(ver217): p.data here is an empty tensor on CUDA and has no useful infomation
# If we change p.grad directly
# it may raise error because of different shape/dtype/device of p.data and p.grad
# We just set p.data = p.colo_attr.saved_grad.payload here
p.data = p.colo_attr.saved_grad.payload
p.grad = p.colo_attr.saved_grad.payload
# Set p.data to empty tensor, in case of memory leaking
p.colo_attr.remove_torch_payload()
def _point_param_fp16_to_master_param(self):
# assign master param pointers to p.data.
# We will not trigger data copy here.
for group in self.optim.param_groups:
for p in group['params']:
self.master_params[p].trans_state(TensorState.COMPUTE)
p.data = self.master_params[p].payload
# Now p.data is sharded
# So optimizer states are sharded naturally
def _copy_master_param_to_param_fp16(self):
# Copy master param data (fp32) to payload of colo_attr (fp16)
# TODO() improve efficiency by gathering tensors into a chunk and transfering
# a chunk.
for group in self.optim.param_groups:
for p in group['params']:
is_param_sharded = p.colo_attr.sharded_data_tensor.is_sharded
if not is_param_sharded and not self.keep_unshard:
# We use ZeRO-2 here
# The `p.colo_attr.sharded_data_tensor` saves full fp16 param
# But we only have updated fp32 param shard here
# So we first shard full fp16 param and copy fp32 param shard to it
# Then we will gather them
self.shard_strategy.shard([p.colo_attr.sharded_data_tensor], self.dp_process_group)
# We have to use `copy_payload` instead of `reset_payload`
# Since p.data is fp32 and p.colo_attr.sharded_data_tensor is fp16
# TODO() optimize this line CPU (fp32) -> GPU (fp16)
p.colo_attr.sharded_data_tensor.reset_payload(
colo_model_tensor_clone(p.half(), torch.cuda.current_device()))
if not is_param_sharded and not self.keep_unshard:
# We gather full fp16 param here
self.shard_strategy.gather([p.colo_attr.sharded_data_tensor], self.dp_process_group)
p.data = p.colo_attr.sharded_data_tensor.payload
self.master_params[p].trans_state(TensorState.HOLD)
p.colo_attr.saved_grad.set_null()