mirror of https://github.com/hpcaitech/ColossalAI
137 lines
6.9 KiB
Python
137 lines
6.9 KiB
Python
from webbrowser import Opera
|
|
import torch
|
|
import torch.nn as nn
|
|
from abc import ABC, abstractmethod
|
|
from torch.fx.node import Node
|
|
from typing import Dict, List
|
|
from colossalai.device.device_mesh import DeviceMesh
|
|
from colossalai.tensor.shape_consistency import ShapeConsistencyManager
|
|
from colossalai.tensor.sharding_spec import ShardingSpec
|
|
|
|
from ..sharding_strategy import StrategiesVector
|
|
|
|
__all__ = ['OperatorHandler']
|
|
|
|
|
|
class OperatorHandler(ABC):
|
|
'''
|
|
The OperatorHandler is an abstract class used to generate every possible strategies for an operator node.
|
|
|
|
Argument:
|
|
input_node(Node): the input node in node argument list.
|
|
input_index(int): the index of input node in the node argument list.
|
|
weight(torch.Tensor): Weight of the node.
|
|
output_node(Node): Output_node is the output of the node.
|
|
device_mesh(DeviceMesh): A logical view of a physical mesh.
|
|
strategies_vector(StrategiesVector): all the strategies generated in this handler will be recorded into the strategies_vector.
|
|
shape_consistency_manager(ShapeConsistencyManager): ShapeConsistencyManager will give the resharding costs of the different sharding specs.
|
|
'''
|
|
|
|
def __init__(self, node: Node, device_mesh: DeviceMesh, strategies_vector: StrategiesVector,
|
|
shape_consistency_manager: ShapeConsistencyManager):
|
|
self.node = node
|
|
self.predecessor_node = list(node._input_nodes.keys())
|
|
self.successor_node = list(node.users.keys())
|
|
self.device_mesh = device_mesh
|
|
self.strategies_vector = strategies_vector
|
|
self.shape_consistency_manager = shape_consistency_manager
|
|
|
|
# find the module and its parameters associated with this node
|
|
# this can be used to compute the compute/communication/sharding cost
|
|
if self.node.op == 'call_module':
|
|
module = node.graph.owning_module.get_submodule(node.target)
|
|
named_parameters = list(module.named_parameters(recurse=False))
|
|
# convert named parameters from list to dict
|
|
named_parameters = {k: v for k, v in named_parameters}
|
|
elif self.node.op == 'call_function':
|
|
module = None
|
|
parameters = list(self.node.args)[1]
|
|
named_parameters = {'weight': parameters._meta_data}
|
|
else:
|
|
module = None
|
|
named_parameters = None
|
|
self.module = module
|
|
self.module_named_parameters = named_parameters
|
|
|
|
@abstractmethod
|
|
def register_strategy(self) -> StrategiesVector:
|
|
"""
|
|
Register
|
|
"""
|
|
pass
|
|
|
|
def _generate_memory_cost(self, dim_partition_dict_for_output, dim_partition_dict_for_weight):
|
|
'''
|
|
Compute the memory cost per device with this specific strategy.
|
|
|
|
Argument:
|
|
dim_partition_dict_for_output(List[int]): The key is the dimension of output to be sharded,
|
|
and the value of the key decribe which logical axis will be sharded in that dimension.
|
|
dim_partition_dict_for_weight(List[int]): The key is the dimension of weight to be sharded,
|
|
and the value of the key decribe which logical axis will be sharded in that dimension.
|
|
Return:
|
|
total_memory_cost(float): total memory cost per device with this specific strategy
|
|
activation_cost(float): the memory cost of activation per device with this specific strategy
|
|
weight_memory_cost(float): the memory cost of weight per device with this specific strategy
|
|
'''
|
|
# compute the size of one element with specific dtype
|
|
dtype = self.input_data.dtype
|
|
size_per_elem_bytes = torch.tensor([], dtype=dtype).element_size()
|
|
|
|
# compute the memory cost of activation
|
|
activation_numel = self.output_data.numel()
|
|
output_mesh_dims = []
|
|
for sharding_dim, mesh_dims in dim_partition_dict_for_output.items():
|
|
output_mesh_dims.extend(mesh_dims)
|
|
activation_sharding_size = 1
|
|
for mesh_dim in output_mesh_dims:
|
|
activation_sharding_size *= self.device_mesh.shape[mesh_dim]
|
|
activation_memory_cost = activation_numel / activation_sharding_size * size_per_elem_bytes
|
|
|
|
# compute the memory cost of weight
|
|
weight_numel = self.weight.numel()
|
|
weight_sharding_size = 1
|
|
weight_mesh_dims = []
|
|
for sharding_dim, mesh_dims in dim_partition_dict_for_weight.items():
|
|
weight_mesh_dims.extend(mesh_dims)
|
|
for mesh_dim in weight_mesh_dims:
|
|
weight_sharding_size *= self.device_mesh.shape[mesh_dim]
|
|
weight_memory_cost = weight_numel / weight_sharding_size * size_per_elem_bytes
|
|
|
|
total_memory_cost = activation_memory_cost + weight_memory_cost
|
|
|
|
return total_memory_cost, activation_memory_cost, weight_memory_cost
|
|
|
|
def _generate_resharding_costs(self, sharding_spec_for_input):
|
|
'''
|
|
Compute the resharding costs with this specific strategy.
|
|
|
|
Note: The resharding_cost of weight is NOT counted.
|
|
|
|
Argument:
|
|
resharding_costs(Dict[int, List[float]]): The resharding cost generated in this method will be appended into this dictionary.
|
|
Resharding_cost[i][j] means the cost of i-th argument in the output node argument list
|
|
with j-th strategy in its strategies_vector transforms to sharding spec wanted in this
|
|
strategy.
|
|
sharding_spec_for_input(ShardingSpec): ShardingSpec of the input node.
|
|
'''
|
|
# The resharding_cost of weight is counted due to sharing weight cases.
|
|
resharding_costs = {}
|
|
dtype = self.node._meta_data.dtype
|
|
size_per_elem_bytes = torch.tensor([], dtype=dtype).element_size()
|
|
for input_node, input_spec in zip(self.predecessor_node, sharding_spec_for_input):
|
|
resharding_costs[input_node] = []
|
|
for strategy in input_node.strategies_vector:
|
|
input_sharding_spec = strategy.output_sharding_spec
|
|
assert isinstance(input_sharding_spec, ShardingSpec), f'The input node should NOT be a tuple of tensor.'
|
|
# compute the resharding cost during forward phase
|
|
_, _, resharding_cost_forward = self.shape_consistency_manager.shape_consistency(
|
|
input_sharding_spec, input_spec)
|
|
# In backward phase, we should convert grad with target_spec into input_sharding_spec
|
|
_, _, resharding_cost_backward = self.shape_consistency_manager.shape_consistency(
|
|
input_spec, input_sharding_spec)
|
|
# we need multiply the size of elem dtype to get correct communication cost
|
|
resharding_cost = (resharding_cost_forward + resharding_cost_backward) * size_per_elem_bytes
|
|
resharding_costs[input_node].append(resharding_cost)
|
|
return resharding_costs
|