ColossalAI/colossalai/zero/low_level/bookkeeping/parameter_store.py

51 lines
1.5 KiB
Python
Raw Normal View History

from torch import Tensor
from torch.distributed import ProcessGroup
from .base_store import BaseStore
class ParameterStore(BaseStore):
def __init__(self, torch_pg: ProcessGroup):
super().__init__(torch_pg)
# record the padding size of each param
self._padding_map = dict()
# mapping working param and master param
self.master_to_working_param = dict()
self.working_to_master_param = dict()
def record_param_padding_size(self, param: Tensor, padding_size: int):
"""Record the padding size of a param
Args:
param (Tensor): The parameter
padding_size (int): The padding size of the parameter
"""
self._padding_map[id(param)] = padding_size
def get_param_padding_size(self, param: Tensor) -> int:
"""Return the padding size of the parameter
Args:
param (Tensor): The parameter
Returns:
int: the padding size of the parameter
"""
return self._padding_map[id(param)]
def link_master_and_working_param(self, master_param: Tensor, working_param: Tensor):
"""Mapping master parameter and working parameter
Args:
master_param (Tensor): The parameter copy in optimizer
working_param (Tensor): The parameter of the model
"""
self.master_to_working_param[id(master_param)] = working_param
self.working_to_master_param[id(working_param)] = master_param