2023-02-18 13:14:38 +00:00
""" This code is adapted from Alpa
https : / / github . com / alpa - projects / alpa /
with some changes . """
2022-08-02 11:23:48 +00:00
import operator
2023-06-15 03:49:16 +00:00
from dataclasses import dataclass
2022-10-27 02:42:54 +00:00
from functools import reduce
2023-06-15 03:49:16 +00:00
from typing import Dict , List , Union
2022-10-27 02:42:54 +00:00
2022-08-02 11:23:48 +00:00
import torch
2022-08-19 05:39:51 +00:00
import torch . distributed as dist
2023-06-15 03:49:16 +00:00
from torch . distributed import ProcessGroup
@dataclass
class ProcessGroupContainer :
process_group : ProcessGroup
ranks : List [ int ]
2022-08-02 11:23:48 +00:00
2023-02-18 13:14:38 +00:00
# modified from alpa LogicalDeviceMesh(https://github.com/alpa-projects/alpa/blob/main/alpa/shard_parallel/auto_sharding.py)
2022-08-02 11:23:48 +00:00
class DeviceMesh :
2023-02-18 13:14:38 +00:00
""" A logical view of a physical cluster. For example, we could view a physical cluster
with 16 devices as a device mesh with shape ( 2 , 2 , 4 ) or ( 4 , 4 ) .
2022-10-27 02:42:54 +00:00
2022-08-02 11:23:48 +00:00
Arguments :
physical_mesh_id ( torch . Tensor ) : physical view of the devices in global rank .
2023-01-11 06:03:49 +00:00
logical_mesh_id ( torch . Tensor ) : logical view of the devices in global rank .
mesh_shape ( torch . Size , optional ) : shape of logical view .
2022-08-02 11:23:48 +00:00
mesh_alpha ( List [ float ] , optional ) : coefficients used for computing
communication cost ( default : None )
mesh_beta ( List [ float ] , optional ) : coefficients used for computing
communication cost ( default : None )
2022-08-19 05:39:51 +00:00
init_process_group ( bool , optional ) : initialize logical process group
during initializing the DeviceMesh instance if the init_process_group set to True .
Otherwise , users need to call create_process_groups_for_logical_mesh manually to init logical process group .
( default : False )
2023-06-15 03:49:16 +00:00
device ( str ) : the device for the process groups used by the DeviceMesh instance . ( default : ' cuda ' )
2022-08-02 11:23:48 +00:00
"""
2023-06-15 03:49:16 +00:00
_DIST_BACKEND = { " cuda " : " nccl " , " cpu " : " gloo " }
2022-08-25 08:48:12 +00:00
def __init__ ( self ,
2023-01-11 06:03:49 +00:00
physical_mesh_id : torch . Tensor ,
mesh_shape : torch . Size = None ,
logical_mesh_id : torch . Tensor = None ,
mesh_alpha : List [ float ] = None ,
mesh_beta : List [ float ] = None ,
init_process_group : bool = False ,
2023-06-15 03:49:16 +00:00
device : str = ' cuda ' ) :
# ============================
# Physical & Logical Mesh IDs
# ============================
self . _physical_mesh_id = physical_mesh_id
assert physical_mesh_id . dim ( ) == 1 , " physical_mesh_id should be a 1D tensor. "
# logical mesh ids can be obtained via two ways
# 1. provide physical mesh id and provide mesh shape
# 2. directly supply the logical mesh id
assert mesh_shape is None or logical_mesh_id is None , \
" Only one of mesh_shape and logical_mesh_id can be specified. " \
" Logical mesh IDs are obtained from either mesh_shape + phyiscal_mesh_id or directly from the user-supplied logical_mesh_id "
2023-01-11 06:03:49 +00:00
if logical_mesh_id is None :
2023-06-15 03:49:16 +00:00
self . _mesh_shape = mesh_shape
self . _logical_mesh_id = self . _physical_mesh_id . reshape ( self . _mesh_shape )
2023-01-11 06:03:49 +00:00
else :
self . _logical_mesh_id = logical_mesh_id
2023-06-15 03:49:16 +00:00
self . _mesh_shape = self . _logical_mesh_id . shape
# ensure two things:
# 1. logical and physical mesh IDs should contain the same elements
# 2. there is no duplicate IDs in each mesh, e.g. [2, 2] is not allowed
assert torch . equal ( torch . unique ( self . _physical_mesh_id ) , torch . unique ( self . logical_mesh_id ) ) , \
" physical and logical mesh IDs should contain the same elements, please check if you have consistent physical_mesh_id and logical_mesh_id. "
assert torch . unique ( self . _physical_mesh_id ) . numel ( ) == self . _physical_mesh_id . numel ( ) , \
" Found duplicate IDs in the phyiscal_mesh_id and this is not allowed, please check your physical_mesh_id again. "
assert torch . unique ( self . logical_mesh_id ) . numel ( ) == self . logical_mesh_id . numel ( ) , \
" Found duplicate IDs in the logical_mesh_id and this is not allowed, please check your logical_mesh_id again. "
2023-01-11 06:03:49 +00:00
2023-06-15 03:49:16 +00:00
# ===============================================
2022-08-02 11:23:48 +00:00
# coefficient for alpha-beta communication model
2023-06-15 03:49:16 +00:00
# alpha is latency and beta is bandwidth
# ===============================================
# if the values are not provided, we assume they are 1 for simplicity
2022-08-02 11:23:48 +00:00
if mesh_alpha is None :
2023-06-15 03:49:16 +00:00
mesh_alpha = [ 1 ] * len ( self . _mesh_shape )
2022-08-02 11:23:48 +00:00
if mesh_beta is None :
2023-06-15 03:49:16 +00:00
mesh_beta = [ 1 ] * len ( self . _mesh_shape )
2022-08-02 11:23:48 +00:00
self . mesh_alpha = tuple ( mesh_alpha )
self . mesh_beta = tuple ( mesh_beta )
2023-06-15 03:49:16 +00:00
# ensure the alpha and beta have the same shape
assert len ( self . mesh_alpha ) == len ( self . mesh_beta ) , \
" mesh_alpha and mesh_beta should have the same length, please check your mesh_alpha and mesh_beta again. "
# =========================
# Device for Process Group
# =========================
self . _device = device
self . _dist_backend = self . _DIST_BACKEND [ device ]
# =========================
# Process Group Management
# =========================
# the _global_to_local_rank_mapping is structured as follows
# {
# <global-rank>: [ <local-rank-on-axis-0>, <local-rank-on-axis-1>, <local-rank-on-axis-2>, ...]
# }
self . _global_to_local_rank_mapping = dict ( )
self . _init_global_to_logical_rank_mapping ( mapping = self . _global_to_local_rank_mapping ,
tensor = self . logical_mesh_id )
# create process group
self . _process_group_dict = { }
self . _ranks_in_the_process_group = { }
self . _global_rank_of_current_process = None
self . _is_initialized = False
# attribute used to inidicate whether this objectd
# is created using DeviceMesh.from_process_group
# this attribute can be used to do some check in methods
# such get_process_group as no global rank information
# is known if created with from_process_group
self . _is_init_from_process_group = False
# initialize process group if specified
self . _init_ranks_in_the_same_group ( )
self . _init_process_group = init_process_group
if init_process_group :
self . init_logical_process_group ( )
2022-08-02 11:23:48 +00:00
@property
2023-06-15 03:49:16 +00:00
def shape ( self ) - > torch . Size :
"""
Return the shape of the logical mesh .
"""
return self . _mesh_shape
2022-08-02 11:23:48 +00:00
@property
2023-06-15 03:49:16 +00:00
def num_devices ( self ) - > int :
"""
Return the number of devices contained in the device mesh .
"""
return reduce ( operator . mul , self . _physical_mesh_id . shape , 1 )
2022-08-02 11:23:48 +00:00
@property
2023-06-15 03:49:16 +00:00
def logical_mesh_id ( self ) - > torch . Tensor :
"""
Return the logical mesh id .
"""
2022-08-02 11:23:48 +00:00
return self . _logical_mesh_id
2023-06-15 03:49:16 +00:00
@property
def is_initialized ( self ) - > bool :
"""
Return whether the process group is initialized .
"""
return self . _is_initialized
@staticmethod
def from_process_group ( process_group : Union [ ProcessGroup , List [ ProcessGroup ] ] ) - > " DeviceMesh " :
"""
Create a DeviceMesh instance from the current process group . Please note that the DeviceMesh object created with this method
will not have information about the physical mesh id , and thus will not be able to query for other ranks and perform alpha - beta communication .
Args :
process_group ( Union [ ProcessGroup , List [ ProcessGroup ] ] ) : the process group or a list of process groups for the device mesh .
If the input is a ProcessGroup object , a 1 D DeviceMesh object will be created . If the input is a list of ProcessGroup objects ,
the ProcessGroup at the ith index will correspond to the process group in the ith axis of the device mesh .
Returns :
DeviceMesh : the device mesh instance .
"""
def _get_device_by_backend ( process_group ) :
"""
Get the device type given a process group ' s backend.
"""
backend = dist . get_backend ( process_group )
for _device , _backend in DeviceMesh . _DIST_BACKEND . items ( ) :
if _backend == backend :
return _device
return None
if isinstance ( process_group , ProcessGroup ) :
process_group = [ process_group ]
# get mesh shape
mesh_shape = [ dist . get_world_size ( pg ) for pg in process_group ]
# get device
device_list = [ _get_device_by_backend ( pg ) for pg in process_group ]
# make sure all devices are the same
assert all ( [ device == device_list [ 0 ] for device in device_list ] ) , \
" All devices should be the same, please check your input process groups are created with the same distributed backend. "
# create a fake physical mesh id
# as we only get the process group associated with the current process,
# we cannot get the global ranks for all processes in the mesh
# therefore, we only use this fake physical mesh id to create the device mesh
# and will remove this fake physical mesh id later
fake_physical_mesh_id = torch . arange ( reduce ( operator . mul , mesh_shape , 1 ) )
# create the device mesh
device_mesh = DeviceMesh ( physical_mesh_id = fake_physical_mesh_id , mesh_shape = mesh_shape , device = device_list [ 0 ] )
# hack the device attribute
device_mesh . _physical_mesh_id = None
device_mesh . _logical_mesh_id = None
device_mesh . _global_rank_of_current_process = dist . get_rank ( )
device_mesh . _is_initialized = False
device_mesh . _process_group_dict = {
device_mesh . _global_rank_of_current_process : { axis : pg for axis , pg in enumerate ( process_group ) }
}
return device_mesh
def get_process_group ( self , axis : int , global_rank : int = None ) - > ProcessGroup :
"""
Return the process group on the specified axis .
Args :
axis ( int ) : the axis of the process group .
global_rank ( int , optional ) : the global rank of the process group . If not specified , the current process is used . ( default : None )
"""
if global_rank is None :
global_rank = self . _global_rank_of_current_process
elif self . _is_init_from_process_group :
raise RuntimeError (
" The logical device mesh is create with DeviceMesh.from_process_group, this method is not supported for this creation method as no global rank information is known. "
)
return self . _process_group_dict [ global_rank ] [ axis ]
def get_process_group_for_all_axes ( self , global_rank : int = None ) - > Dict [ int , ProcessGroup ] :
"""
Return the process groups for all axes .
Args :
global_rank ( int , optional ) : the global rank of the process
"""
if global_rank is None :
global_rank = self . _global_rank_of_current_process
elif self . _is_init_from_process_group :
raise RuntimeError (
" The logical device mesh is create with DeviceMesh.from_process_group, this method is not supported for this creation method as no global rank information is known. "
)
return self . _process_group_dict [ global_rank ]
def get_ranks_in_process_group ( self , axis : int , global_rank : int = None ) - > List [ int ] :
"""
Return the ranks in the process group on the specified axis .
Args :
axis ( int ) : the axis of the process group .
global_rank ( int , optional ) : the global rank of the process
"""
if global_rank is None :
global_rank = self . _global_rank_of_current_process
elif self . _is_init_from_process_group :
raise RuntimeError (
" The logical device mesh is create with DeviceMesh.from_process_group, this method is not supported for this creation method as no global rank information is known. "
)
return self . _ranks_in_the_process_group [ global_rank ] [ axis ]
def __deepcopy__ ( self , memo ) - > " DeviceMesh " :
2022-10-27 02:42:54 +00:00
cls = self . __class__
result = cls . __new__ ( cls )
memo [ id ( self ) ] = result
for k , v in self . __dict__ . items ( ) :
2023-06-15 03:49:16 +00:00
if k != ' _process_group_dict ' :
2022-10-27 02:42:54 +00:00
setattr ( result , k , __import__ ( " copy " ) . deepcopy ( v , memo ) )
else :
2023-06-15 03:49:16 +00:00
# process group cannot be copied
# thus, we share them directly
2022-10-27 02:42:54 +00:00
setattr ( result , k , v )
return result
2023-06-15 03:49:16 +00:00
def _init_global_to_logical_rank_mapping ( self ,
mapping : Dict ,
tensor : torch . Tensor ,
index_list : List [ int ] = [ ] ) - > Dict [ int , List [ int ] ] :
2023-06-08 02:18:17 +00:00
"""
2023-06-15 03:49:16 +00:00
Build a global rank to local rank mapping for each process group in different axis in the logical device mesh .
2023-06-08 02:18:17 +00:00
2023-06-15 03:49:16 +00:00
Args :
mapping ( Dict ) : a dictionary that maps the global rank to the local rank in the logical device mesh .
tensor ( torch . Tensor ) : the tensor that contains the logical mesh ids .
index_list ( List [ int ] )
Returns :
mapping ( Dict ) : a dictionary that maps the global rank to the local rank in the logical device mesh .
The value is a list of integers and each integer represents the local rank in the indexed axis .
"""
2023-06-09 01:41:27 +00:00
for index , inner_tensor in enumerate ( tensor ) :
2023-06-15 03:49:16 +00:00
# index means the local rank in the current axis
# inner_tensor refers to the processes with the same local rank
2022-08-02 11:23:48 +00:00
if inner_tensor . numel ( ) == 1 :
2023-06-15 03:49:16 +00:00
# if the inner_tensor only has one element, it means that
# it already reaches the last axis
# we append its local_rank in the last axis to the index_list
# and assign to the mapping
# the value of the mapping is the the local rank at the indexed axis of the device mesh
mapping [ int ( inner_tensor ) ] = index_list + [ index ]
2022-08-02 11:23:48 +00:00
else :
2023-06-15 03:49:16 +00:00
# we recursively go into the function until we reach the last axis
# meanwhile, we should add the local rank in the current axis in the index_list
self . _init_global_to_logical_rank_mapping ( mapping , inner_tensor , index_list + [ index ] )
2022-08-02 11:23:48 +00:00
2023-06-15 03:49:16 +00:00
def init_logical_process_group ( self ) :
2022-08-19 05:39:51 +00:00
'''
This method is used to initialize the logical process groups which will be used in communications
2022-10-27 02:42:54 +00:00
among logical device mesh .
2022-08-19 05:39:51 +00:00
Note : if init_process_group set to False , you have to call this method manually . Otherwise ,
the communication related function , such as ShapeConsistencyManager . apply will raise errors .
'''
2023-06-15 03:49:16 +00:00
# sanity check
assert dist . is_initialized , " The torch.distributed should be initialized before calling init_logical_process_group "
assert not self . _is_initialized , " The logical process group has been initialized, do not call init_logical_process_group twice "
# update the global rank of the current process
self . _global_rank_of_current_process = dist . get_rank ( )
duplicate_check_list = [ ]
# flatten the global ranks to 1D list
global_rank_flatten_list = self . _physical_mesh_id . view ( - 1 ) . tolist ( )
2023-06-08 02:18:17 +00:00
for global_rank in global_rank_flatten_list :
2023-06-15 03:49:16 +00:00
# find the other ranks which are in the same process group as global_rank
ranks_in_same_group_by_axis = self . _collate_global_ranks_in_same_process_group ( global_rank )
2023-06-08 02:18:17 +00:00
2023-06-15 03:49:16 +00:00
for axis , ranks_in_same_group in ranks_in_same_group_by_axis . items ( ) :
# skip duplicated process group creation
if ranks_in_same_group in duplicate_check_list :
continue
2023-06-08 02:18:17 +00:00
2023-06-15 03:49:16 +00:00
# create the process group
pg_handler = dist . new_group ( ranks = ranks_in_same_group , backend = self . _dist_backend )
2023-06-08 02:18:17 +00:00
2023-06-15 03:49:16 +00:00
# keep this process group in the process_groups_dict
for rank in ranks_in_same_group :
if rank not in self . _process_group_dict :
self . _process_group_dict [ rank ] = dict ( )
self . _process_group_dict [ rank ] [ axis ] = pg_handler
# update the init flag
# we only allow init for once
self . _is_initialized = True
def _init_ranks_in_the_same_group ( self ) :
"""
This method is used to initialize the ranks_in_the_same_group dictionary .
"""
# flatten the global ranks to 1D list
global_rank_flatten_list = self . _physical_mesh_id . view ( - 1 ) . tolist ( )
for global_rank in global_rank_flatten_list :
# find the other ranks which are in the same process group as global_rank
ranks_in_same_group_by_axis = self . _collate_global_ranks_in_same_process_group ( global_rank )
for axis , ranks_in_same_group in ranks_in_same_group_by_axis . items ( ) :
# create dict for each rank
if global_rank not in self . _process_group_dict :
self . _ranks_in_the_process_group [ global_rank ] = dict ( )
# keep this process group in the process_groups_dict
self . _ranks_in_the_process_group [ global_rank ] [ axis ] = ranks_in_same_group
def global_rank_to_local_rank ( self , rank : int , axis : int = None ) - > Union [ List [ int ] , int ] :
"""
Return the local rank of the given global rank in the logical device mesh .
Args :
rank ( int ) : the global rank in the logical device mesh .
axis ( int ) : the axis of the logical device mesh .
"""
if self . _is_init_from_process_group :
raise RuntimeError (
" The logical device mesh is create with DeviceMesh.from_process_group, this method is not supported for this creation method as no global rank information is known. "
)
local_ranks = self . _global_to_local_rank_mapping [ rank ]
if axis :
return local_ranks [ axis ]
else :
return local_ranks
def _collate_global_ranks_in_same_process_group ( self , global_rank ) :
2023-06-09 01:41:27 +00:00
'''
2023-06-15 03:49:16 +00:00
Give a global rank and return all global ranks involved in its associated process group in each axis .
Example :
` ` ` python
sphysical_mesh_id = torch . arange ( 0 , 16 )
mesh_shape = ( 4 , 4 )
# logical mesh will look like
# [[0, 1, 2, 3],
# [4, 5, 6, 7],
# [8, 9, 10,11],
# [12,13,14,15]]
device_mesh = DeviceMesh ( physical_mesh_id , mesh_shape )
print ( device_mesh . collate_global_ranks_in_same_process_group ( 0 ) )
# key is axis name
# value is a list of global ranks in same axis with rank 0
# output will look like
# {
0 : [ 0 , 4 , 8 , 12 ] ,
1 : [ 0 , 1 , 2 , 3 ]
# }
2023-06-09 01:41:27 +00:00
'''
2023-06-15 03:49:16 +00:00
# We have init the global rank to local rank by calling _init_global_to_logical_rank_mapping
# for self._global_to_local_rank_mapping
# the key is the global rank
# the value is the list of local ranks corresponding to the global rank with respect of different axes
# we can see the list of local ranks as the process coordinates for simplicity
# the key and value are all unique, therefore,
# we can also to use the coordinates to find the global rank
# =========================================================================
# Step 1
# find all the process_coordinates for processes in the same process group
# as the given global rank
# =========================================================================
# each
processes_in_the_same_process_group = { }
for dim in range ( self . logical_mesh_id . dim ( ) ) :
# iterate over the dimension size so that we can include all processes
# in the same process group in the given axis
# the _local_rank refers to the local rank of the current process
for _local_rank in range ( self . logical_mesh_id . shape [ dim ] ) :
# if this dimension is not initailized yet,
# initialize it with an empty array
if dim not in processes_in_the_same_process_group :
processes_in_the_same_process_group [ dim ] = [ ]
# get the local rank corresponding to the global rank
process_coordinates = self . _global_to_local_rank_mapping [ global_rank ] . copy ( )
# replace the local rank in the given dimension with the
# lcoal rank of the current process iterated
process_coordinates [ dim ] = _local_rank
processes_in_the_same_process_group [ dim ] . append ( process_coordinates )
# =================================================================
# Step 2
# Use local rank combination to find its corresponding global rank
# =================================================================
# the key of the dict is the axis
# the value is the list of global ranks which are in the same process group as the given global rank
global_pg_ranks = { }
for dim , coordinates_of_all_processes in processes_in_the_same_process_group . items ( ) :
global_pg_ranks [ dim ] = [ ]
for process_coordinates in coordinates_of_all_processes :
# find the global rank by local rank combination
for _global_rank , _process_coordinates in self . _global_to_local_rank_mapping . items ( ) :
if process_coordinates == _process_coordinates :
global_pg_ranks [ dim ] . append ( _global_rank )
return global_pg_ranks
def flatten ( self ) :
"""
Flatten the logical mesh into an effective 1 d logical mesh ,
"""
if self . _is_init_from_process_group :
raise RuntimeError (
" The logical device mesh is create with DeviceMesh.from_process_group, this method is not supported for this creation method as no global rank information is known. "
)
flatten_mesh_shape_size = len ( self . _mesh_shape )
flatten_mesh_shape = [ self . num_devices ]
return DeviceMesh ( self . _physical_mesh_id ,
tuple ( flatten_mesh_shape ) ,
mesh_alpha = [ max ( self . mesh_alpha ) ] * ( flatten_mesh_shape_size - 1 ) ,
mesh_beta = [ max ( self . mesh_beta ) ] * ( flatten_mesh_shape_size - 1 ) ,
init_process_group = self . _init_process_group )
2022-08-02 11:23:48 +00:00
def all_gather_cost ( self , num_bytes , mesh_dim ) :
num_devices = self . logical_mesh_id . shape [ mesh_dim ]
return ( self . mesh_alpha [ mesh_dim ] + self . mesh_beta [ mesh_dim ] * ( num_devices - 1 ) / num_devices * num_bytes +
0.1 )
def all_reduce_cost ( self , num_bytes , mesh_dim ) :
num_devices = self . logical_mesh_id . shape [ mesh_dim ]
return ( self . mesh_alpha [ mesh_dim ] + self . mesh_beta [ mesh_dim ] * 2 * ( num_devices - 1 ) / num_devices * num_bytes +
0.01 )
def reduce_scatter_cost ( self , num_bytes , mesh_dim ) :
num_devices = self . logical_mesh_id . shape [ mesh_dim ]
return ( self . mesh_alpha [ mesh_dim ] + self . mesh_beta [ mesh_dim ] * ( num_devices - 1 ) / num_devices * num_bytes +
0.001 )
def all_to_all_cost ( self , num_bytes , mesh_dim ) :
num_devices = self . logical_mesh_id . shape [ mesh_dim ]
penalty_factor = num_devices / 2.0
return ( self . mesh_alpha [ mesh_dim ] + self . mesh_beta [ mesh_dim ] *
2023-06-15 03:49:16 +00:00
( num_devices - 1 ) / num_devices / num_devices * num_bytes * penalty_factor + 0.001 )