github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-09-04 18:40:28 +00:00
Code Issues Packages Projects Releases Wiki Activity

[misc] update pre-commit and run all files (#4752)

Browse Source 
* [misc] update pre-commit

* [misc] run pre-commit

* [misc] remove useless configuration files

* [misc] ignore cuda for clang-format
This commit is contained in:
Hongxin Liu
2023-09-19 14:20:26 +08:00
committed by GitHub
parent 3c6b831c26
commit 079bf3cb26
1268 changed files with 50037 additions and 38444 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
colossalai/legacy/context/process_group_initializer/__init__.py
View File

@@ -10,6 +10,14 @@ from .initializer_tensor import Initializer_Tensor
from .process_group_initializer import ProcessGroupInitializer
__all__ = [
'Initializer_Tensor', 'Initializer_Sequence', 'Initializer_Pipeline', 'Initializer_Data', 'Initializer_2p5D',
'Initializer_2D', 'Initializer_3D', 'Initializer_1D', 'ProcessGroupInitializer', 'Initializer_Model'
"Initializer_Tensor",
"Initializer_Sequence",
"Initializer_Pipeline",
"Initializer_Data",
"Initializer_2p5D",
"Initializer_2D",
"Initializer_3D",
"Initializer_1D",
"ProcessGroupInitializer",
"Initializer_Model",
]

2
colossalai/legacy/context/process_group_initializer/initializer_1d.py
View File

@@ -45,7 +45,7 @@ class Initializer_1D(ProcessGroupInitializer):
for i in range(self.num_group):
ranks = [i * self.tensor_parallel_size + j for j in range(self.tensor_parallel_size)]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)

16
colossalai/legacy/context/process_group_initializer/initializer_2d.py
View File

@@ -14,9 +14,10 @@ def _check_summa_env_var(summa_dim):
env_summa_dim = env.summa_dim
if env_summa_dim:
assert int(env_summa_dim) == summa_dim, \
'SUMMA_DIM has been set in the current environment and ' \
'does not match with the value passed to this initialized'
assert int(env_summa_dim) == summa_dim, (
"SUMMA_DIM has been set in the current environment and "
"does not match with the value passed to this initialized"
)
else:
env.summa_dim = summa_dim
@@ -57,7 +58,7 @@ class Initializer_2D_Row(ProcessGroupInitializer):
for j in range(self.summa_dim):
ranks = [i * self.tensor_parallel_size + j * self.summa_dim + k for k in range(self.summa_dim)]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)
@@ -106,7 +107,7 @@ class Initializer_2D_Col(ProcessGroupInitializer):
for j in range(self.summa_dim):
ranks = [i * self.tensor_parallel_size + j + k * self.summa_dim for k in range(self.summa_dim)]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)
@@ -137,8 +138,9 @@ class Initializer_2D(ProcessGroupInitializer):
self.num_group = self.world_size // self.tensor_parallel_size
self.summa_dim = int(math.sqrt(self.tensor_parallel_size))
assert self.tensor_parallel_size == self.summa_dim ** 2, \
"2D summa dim should equal to tensor parallel size ^ 0.5"
assert (
self.tensor_parallel_size == self.summa_dim**2
), "2D summa dim should equal to tensor parallel size ^ 0.5"
_check_summa_env_var(self.summa_dim)
self.col_initializer = Initializer_2D_Col(self.num_group, self.summa_dim, *args, **kwargs)

46
colossalai/legacy/context/process_group_initializer/initializer_2p5d.py
View File

@@ -19,12 +19,14 @@ def _check_tesseract_env_var(tesseract_dim: int, tesseract_dep: int):
env_tesseract_dep = env.tesseract_dep
if env_tesseract_dim and env_tesseract_dep:
assert int(env_tesseract_dim) == tesseract_dim, \
'TESSERACT_DIM has been set in the current environment and ' \
'does not match with the value passed to this initialized'
assert int(env_tesseract_dep) == tesseract_dep, \
'TESSERACT_DEP has been set in the current environment and ' \
'does not match with the value passed to this initialized'
assert int(env_tesseract_dim) == tesseract_dim, (
"TESSERACT_DIM has been set in the current environment and "
"does not match with the value passed to this initialized"
)
assert int(env_tesseract_dep) == tesseract_dep, (
"TESSERACT_DEP has been set in the current environment and "
"does not match with the value passed to this initialized"
)
else:
env.tesseract_dim = tesseract_dim
env.tesseract_dep = tesseract_dep
@@ -50,8 +52,9 @@ class Initializer_2p5D_ROW(ProcessGroupInitializer):
self.num_group = self.world_size // self.tensor_parallel_size
self.tesseract_dep = tesseract_dep
self.tesseract_dim = tesseract_dim
assert self.tensor_parallel_size == self.tesseract_dim ** 2 * self.tesseract_dep, \
"Tensor parallel size should be depth * dim ** 2 in 2.5D parallel"
assert (
self.tensor_parallel_size == self.tesseract_dim**2 * self.tesseract_dep
), "Tensor parallel size should be depth * dim ** 2 in 2.5D parallel"
def init_dist_group(self):
"""Initialize 2.5D tensor row parallel groups, and assign local_ranks and groups to each gpu.
@@ -75,7 +78,7 @@ class Initializer_2p5D_ROW(ProcessGroupInitializer):
for i in range(self.tesseract_dim)
]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)
@@ -129,7 +132,7 @@ class Initializer_2p5D_Col(ProcessGroupInitializer):
for j in range(self.tesseract_dim)
]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)
@@ -183,7 +186,7 @@ class Initializer_2p5D_Dep(ProcessGroupInitializer):
for k in range(self.tesseract_dep)
]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)
@@ -238,7 +241,7 @@ class Initializer_2p5D_XZ(ProcessGroupInitializer):
for j in range(self.tesseract_dim)
]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)
@@ -265,16 +268,25 @@ class Initializer_2p5D(ProcessGroupInitializer):
depth (int): The depth of 2.5d parallel.
"""
def __init__(self, rank: int, world_size: int, config: Config, data_parallel_size: int, pipeline_parallel_size: int,
tensor_parallel_size: int, depth: int):
def __init__(
self,
rank: int,
world_size: int,
config: Config,
data_parallel_size: int,
pipeline_parallel_size: int,
tensor_parallel_size: int,
depth: int,
):
args = (rank, world_size, config, data_parallel_size, pipeline_parallel_size, tensor_parallel_size)
super().__init__(*args)
self.num_group = self.world_size // self.tensor_parallel_size
self.tesseract_dim = int(math.sqrt(self.tensor_parallel_size / depth))
self.tesseract_dep = depth
assert self.tensor_parallel_size == self.tesseract_dim ** 2 * self.tesseract_dep, \
"2.5D tesseract dim should equal to (tensor parallel size / tesseract dep) ^ 0.5"
assert (
self.tensor_parallel_size == self.tesseract_dim**2 * self.tesseract_dep
), "2.5D tesseract dim should equal to (tensor parallel size / tesseract dep) ^ 0.5"
_check_tesseract_env_var(self.tesseract_dim, self.tesseract_dep)
self.col_initializer = Initializer_2p5D_Col(self.tesseract_dim, self.tesseract_dep, *args)
@@ -293,6 +305,6 @@ class Initializer_2p5D(ProcessGroupInitializer):
self.col_initializer.init_dist_group(),
self.row_initializer.init_dist_group(),
self.dep_initializer.init_dist_group(),
self.xz_initializer.init_dist_group()
self.xz_initializer.init_dist_group(),
]
return parallel_setting

24
colossalai/legacy/context/process_group_initializer/initializer_3d.py
View File

@@ -17,9 +17,10 @@ def _check_depth_env_var(depth):
env_depth = env.depth_3d
if env_depth:
assert int(env_depth) == depth, \
'DEPTH_3D has been set in the current environment and ' \
'does not match with the value passed to this initialized'
assert int(env_depth) == depth, (
"DEPTH_3D has been set in the current environment and "
"does not match with the value passed to this initialized"
)
else:
env.depth_3d = depth
@@ -63,7 +64,7 @@ class Initializer_3D_Input(ProcessGroupInitializer):
for k in range(self.depth):
ranks = [h * self.depth**3 + i + self.depth * (j + self.depth * k) for j in range(self.depth)]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)
@@ -114,7 +115,7 @@ class Initializer_3D_Weight(ProcessGroupInitializer):
for j in range(self.depth):
ranks = [h * self.depth**3 + i + self.depth * (j + self.depth * k) for i in range(self.depth)]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)
@@ -165,7 +166,7 @@ class Initializer_3D_Output(ProcessGroupInitializer):
for j in range(self.depth):
ranks = [h * self.depth**3 + i + self.depth * (j + self.depth * k) for k in range(self.depth)]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)
@@ -219,7 +220,7 @@ class Initializer_3D_InputxWeight(ProcessGroupInitializer):
for i in range(self.depth)
]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)
@@ -273,7 +274,7 @@ class Initializer_3D_OutputxWeight(ProcessGroupInitializer):
for i in range(self.depth)
]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)
@@ -302,8 +303,9 @@ class Initializer_3D(ProcessGroupInitializer):
super().__init__(*args)
self.num_group = self.world_size // self.tensor_parallel_size
self.depth = round(math.pow(self.tensor_parallel_size, 1 / 3))
assert self.tensor_parallel_size == self.depth ** 3, \
f'3D depth ({self.depth}) if not cube root of tensor parallel size ({self.tensor_parallel_size})'
assert (
self.tensor_parallel_size == self.depth**3
), f"3D depth ({self.depth}) if not cube root of tensor parallel size ({self.tensor_parallel_size})"
_check_depth_env_var(self.depth)
self.input_initializer = Initializer_3D_Input(self.num_group, self.depth, *args)
@@ -324,6 +326,6 @@ class Initializer_3D(ProcessGroupInitializer):
self.weight_initializer.init_dist_group(),
self.output_initializer.init_dist_group(),
self.input_x_weight_initializer.init_dist_group(),
self.output_x_weight_initializer.init_dist_group()
self.output_x_weight_initializer.init_dist_group(),
]
return parallel_setting

2
colossalai/legacy/context/process_group_initializer/initializer_data.py
View File

@@ -43,7 +43,7 @@ class Initializer_Data(ProcessGroupInitializer):
for i in range(self.num_data_parallel_group):
ranks = [i + j * self.num_data_parallel_group for j in range(self.data_parallel_size)]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)

2
colossalai/legacy/context/process_group_initializer/initializer_model.py
View File

@@ -45,7 +45,7 @@ class Initializer_Model(ProcessGroupInitializer):
for i in range(self.num_group):
ranks = [i * self.model_parallel_size + j for j in range(self.model_parallel_size)]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)

18
colossalai/legacy/context/process_group_initializer/initializer_pipeline.py
View File

@@ -38,10 +38,11 @@ class Initializer_Pipeline(ProcessGroupInitializer):
for i in range(self.data_parallel_size):
for j in range(self.pipeline_stage_size):
pipe_ranks = list(
range(i * self.data_group_size + j, (i + 1) * self.data_group_size, self.pipeline_stage_size))
range(i * self.data_group_size + j, (i + 1) * self.data_group_size, self.pipeline_stage_size)
)
pipe_group_size = len(pipe_ranks)
pipe_group = dist.new_group(pipe_ranks)
group_cpu = dist.new_group(pipe_ranks, backend='gloo') if dist.get_backend() != 'gloo' else pipe_group
group_cpu = dist.new_group(pipe_ranks, backend="gloo") if dist.get_backend() != "gloo" else pipe_group
if self.rank in pipe_ranks:
local_rank = pipe_ranks.index(self.rank)
@@ -50,7 +51,16 @@ class Initializer_Pipeline(ProcessGroupInitializer):
cpu_group = group_cpu
ranks_in_group = pipe_ranks
dist_settings.append(
tuple((local_rank, group_world_size, process_group, cpu_group, ranks_in_group,
ParallelMode.PIPELINE)))
tuple(
(
local_rank,
group_world_size,
process_group,
cpu_group,
ranks_in_group,
ParallelMode.PIPELINE,
)
)
)
return dist_settings

12
colossalai/legacy/context/process_group_initializer/initializer_sequence.py
View File

@@ -46,7 +46,7 @@ class Initializer_Sequence_DP(ProcessGroupInitializer):
for i in range(self.num_group):
ranks = [i * self.dp_size + j for j in range(self.dp_size)]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)
@@ -91,8 +91,14 @@ class Initializer_Sequence(ProcessGroupInitializer):
parallel_setting = []
local_rank, group_world_size, process_group, cpu_group, ranks_in_group, mode = \
self._sequence_initializer.init_dist_group()
(
local_rank,
group_world_size,
process_group,
cpu_group,
ranks_in_group,
mode,
) = self._sequence_initializer.init_dist_group()
# change mode to sequence
mode = ParallelMode.SEQUENCE

2
colossalai/legacy/context/process_group_initializer/initializer_tensor.py
View File

@@ -43,7 +43,7 @@ class Initializer_Tensor(ProcessGroupInitializer):
for i in range(self.num_tensor_parallel_group):
ranks = [i * self.tensor_parallel_size + j for j in range(self.tensor_parallel_size)]
group = dist.new_group(ranks)
group_cpu = dist.new_group(ranks, backend='gloo') if dist.get_backend() != 'gloo' else group
group_cpu = dist.new_group(ranks, backend="gloo") if dist.get_backend() != "gloo" else group
if self.rank in ranks:
local_rank = ranks.index(self.rank)

11
colossalai/legacy/context/process_group_initializer/process_group_initializer.py
View File

@@ -18,8 +18,15 @@ class ProcessGroupInitializer(ABC):
tensor_parallel_size (int): Size of tensor parallel.
"""
def __init__(self, rank: int, world_size: int, config: Config, data_parallel_size: int, pipeline_parallel_size: int,
tensor_parallel_size: int):
def __init__(
self,
rank: int,
world_size: int,
config: Config,
data_parallel_size: int,
pipeline_parallel_size: int,
tensor_parallel_size: int,
):
self.rank = rank
self.world_size = world_size
self.data_parallel_size = data_parallel_size