[zero] find miss code (#378)

pull/394/head
Jiarui Fang 3 years ago committed by Frank Lee
parent 6b6002962a
commit b5f43acee3

@ -4,10 +4,6 @@ repos:
hooks: hooks:
- id: yapf - id: yapf
args: ['--style=.style.yapf', '--parallel', '--in-place'] args: ['--style=.style.yapf', '--parallel', '--in-place']
- repo: https://github.com/pycqa/flake8
rev: '4.0.1'
hooks:
- id: flake8
- repo: https://github.com/pre-commit/mirrors-clang-format - repo: https://github.com/pre-commit/mirrors-clang-format
rev: v13.0.1 rev: v13.0.1
hooks: hooks:

@ -0,0 +1,3 @@
from .bucket_tensor_copy import BucketizedTensorCopy
__all__ = ['BucketizedTensorCopy']

@ -0,0 +1,61 @@
import torch
from colossalai.zero.sharded_param import ShardedParamV2
from colossalai.utils import get_current_device
from typing import List
class BucketizedTensorCopy(object):
def __init__(
self,
chunk_size: int,
):
r"""
torch.nn.Parameter CPU (fp32) -> ShardedParam GPU (fp16)
TODO(jiaruifang) The class is a little bit hardcoded
I will make it more general later.
"""
self.chunk_size = chunk_size
self._offset = 0
self._cpu_buffer = torch.empty(chunk_size, dtype=torch.float, device=torch.device("cpu:0"), pin_memory=True)
self._cuda_buffer = torch.empty(chunk_size,
dtype=torch.half,
device=torch.device(f"cuda:{get_current_device()}"))
self._buffered_param_list: List[ShardedParamV2] = []
self._numel_list = []
def copy(self, src_param: torch.nn.Parameter, target_param: ShardedParamV2):
assert isinstance(target_param, ShardedParamV2)
assert isinstance(src_param, torch.nn.Parameter)
numel = src_param.numel()
if self._offset + numel > self.chunk_size:
self.flush()
assert src_param.data.device.type == 'cpu'
self._cpu_buffer.narrow(0, self._offset, numel).copy_(src_param.data.view(-1))
self._buffered_param_list.append(target_param)
self._numel_list.append(numel)
self._offset += numel
def flush(self):
"""
flush to cuda memory
"""
self._cuda_buffer.copy_(self._cpu_buffer)
flush_offset = 0
for sparam, numel in zip(self._buffered_param_list, self._numel_list):
sparam.data.copy_payload(self._cpu_buffer.narrow(0, flush_offset, numel))
flush_offset += numel
self.reset()
def reset(self):
self._buffered_param_list = []
self._numel_list = []
self._offset = 0

@ -88,14 +88,19 @@ class ShardedOptimizerV2(ColossalaiOptimizer):
self.zero_grad() self.zero_grad()
return return
# Write master param to p.data # assign master param pointers to p.data.
# We will not trigger data copy here.
for group in self.optim.param_groups: for group in self.optim.param_groups:
for p in group['params']: for p in group['params']:
p.data = self.master_params[p] p.data = self.master_params[p]
# Now p.data is sharded # Now p.data is sharded
# So optimizer states are sharded naturally # So optimizer states are sharded naturally
ret = self.optim.step(*args, **kwargs) ret = self.optim.step(*args, **kwargs)
# Write master param to payload
# Copy master param data (fp32) to payload of col_attr (fp16)
# TODO() improve efficiency by gathering tensors into a chunk and transfering
# a chunk.
for group in self.optim.param_groups: for group in self.optim.param_groups:
for p in group['params']: for p in group['params']:
is_param_sharded = p.col_attr.data.is_sharded is_param_sharded = p.col_attr.data.is_sharded
@ -108,7 +113,10 @@ class ShardedOptimizerV2(ColossalaiOptimizer):
self.shard_strategy.shard([p.col_attr.data]) self.shard_strategy.shard([p.col_attr.data])
# We have to use `copy_payload` instead of `reset_payload` # We have to use `copy_payload` instead of `reset_payload`
# Since p.data is fp32 and p.col_attr.data is fp16 # Since p.data is fp32 and p.col_attr.data is fp16
# TODO() optimize this line
p.col_attr.data.copy_payload(p.data) p.col_attr.data.copy_payload(p.data)
if not is_param_sharded: if not is_param_sharded:
# We gather full fp16 param here # We gather full fp16 param here
self.shard_strategy.gather([p.col_attr.data]) self.shard_strategy.gather([p.col_attr.data])

@ -14,7 +14,6 @@ class ShardedTensor(object):
self.world_size = dist.get_world_size(self.process_group) self.world_size = dist.get_world_size(self.process_group)
self.local_rank = dist.get_rank(self.process_group) self.local_rank = dist.get_rank(self.process_group)
self._is_sharded = False self._is_sharded = False
self._payload = tensor
self._origin_shape = tensor.shape self._origin_shape = tensor.shape
self._origin_numel = tensor.numel() self._origin_numel = tensor.numel()
@ -41,7 +40,7 @@ class ShardedTensor(object):
return self._payload return self._payload
def copy_payload(self, tensor): def copy_payload(self, tensor):
self._payload.copy_(tensor) self._payload.view(-1).copy_(tensor.view(-1))
def reset_payload(self, tensor): def reset_payload(self, tensor):
del self._payload del self._payload

@ -0,0 +1,39 @@
from colossalai.utils.commons import BucketizedTensorCopy
from colossalai.zero.sharded_param import ShardedParamV2
from colossalai.utils import free_port
import torch
import colossalai
def test_bucket_copy():
# init dist env
colossalai.launch(config={}, rank=0, world_size=1, host='localhost', port=free_port(), backend='nccl')
copyer = BucketizedTensorCopy(20)
shape_list = [(2, 3), (5), (8), (12)]
src_param_list = []
tgt_param_list = []
for shape in shape_list:
# on CPU
src_param = torch.nn.Parameter(torch.randn(shape, dtype=torch.float, device=torch.device('cpu')))
print(src_param)
# on GPU
tgt_param = ShardedParamV2(torch.nn.Parameter(torch.ones(shape, dtype=torch.half, device=torch.device('cuda'))))
src_param_list.append(src_param)
tgt_param_list.append(tgt_param)
copyer.copy(src_param, tgt_param)
copyer.flush()
for src_param, tgt_param in zip(src_param_list, tgt_param_list):
print(tgt_param.data.payload)
diff = src_param.cpu().float() - tgt_param.data.payload.cpu().float()
assert torch.allclose(src_param.cpu().float(), tgt_param.data.payload.cpu().float(), rtol=1e-03,
atol=1e-03), f"diff {diff}"
if __name__ == '__main__':
test_bucket_copy()
Loading…
Cancel
Save