mirror of https://github.com/hpcaitech/ColossalAI
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
62 lines
1.9 KiB
62 lines
1.9 KiB
3 years ago
|
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):
|
||
3 years ago
|
sparam.sharded_data_tensor.copy_payload(self._cpu_buffer.narrow(0, flush_offset, numel))
|
||
3 years ago
|
flush_offset += numel
|
||
|
|
||
|
self.reset()
|
||
|
|
||
|
def reset(self):
|
||
|
self._buffered_param_list = []
|
||
|
self._numel_list = []
|
||
|
self._offset = 0
|