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.
474 lines
20 KiB
474 lines
20 KiB
1 year ago
|
#!/usr/bin/env python
|
||
|
# -*- encoding: utf-8 -*-
|
||
|
|
||
|
import math
|
||
|
from typing import Callable, List, Tuple, Union
|
||
|
|
||
|
import torch
|
||
|
import torch.distributed as dist
|
||
|
import torch.nn as nn
|
||
|
import torch.nn.functional as F
|
||
|
from torch import Tensor
|
||
|
from torch.distributed import ProcessGroup
|
||
|
from torch.nn.parameter import Parameter
|
||
|
|
||
|
from colossalai.nn import init as init
|
||
|
from colossalai.nn.layer.utils import divide
|
||
1 year ago
|
from colossalai.tensor.d_tensor.api import (
|
||
|
customized_distributed_tensor_to_param,
|
||
|
distribute_tensor_with_customization,
|
||
|
shard_rowwise,
|
||
|
sharded_tensor_to_param,
|
||
|
)
|
||
1 year ago
|
|
||
|
from ._operation import (
|
||
|
gather_forward_split_backward,
|
||
1 year ago
|
matmul_with_async_comm,
|
||
|
reduce_backward,
|
||
|
reduce_forward,
|
||
1 year ago
|
split_forward_gather_backward,
|
||
|
)
|
||
1 year ago
|
from .parallel_module import ParallelModule
|
||
1 year ago
|
from .utils import create_randomizer_with_offset
|
||
|
|
||
1 year ago
|
__all__ = ['FusedLinear1D_Col', 'FusedLinear1D_Row']
|
||
1 year ago
|
|
||
1 year ago
|
# ====================================
|
||
|
# For GPT Only
|
||
|
# ====================================
|
||
1 year ago
|
|
||
1 year ago
|
|
||
|
def split_fused_qkv_in_gpt2_style(qkv: torch.Tensor,
|
||
|
n_fused: int,
|
||
|
process_group: ProcessGroup,
|
||
|
is_transposed: bool = False):
|
||
1 year ago
|
"""
|
||
|
The fused qkv tensor looks like [Q1, Q2, K1, K2, V1, V2], this function will split them into [Q1, K1, V1] and [Q2, K2, V2].
|
||
1 year ago
|
|
||
|
Args:
|
||
|
qkv (torch.Tensor): The fused qkv tensor.
|
||
|
n_fused (int): The number items fused together, defaults to 3 (query, key and value).
|
||
|
process_group (ProcessGroup): The process group for distributed communication.
|
||
|
is_transposed (bool): generally the tensor is the shape of (out_features, in_features). Set this to True if the tensor is in the shape (in_features, out_features).
|
||
1 year ago
|
"""
|
||
|
# get the number of slice for the fused qkv
|
||
|
rank = dist.get_rank(group=process_group)
|
||
|
world_size = dist.get_world_size(group=process_group)
|
||
|
order = torch.arange(world_size * n_fused)
|
||
|
|
||
|
# split the fused qkv
|
||
|
# from
|
||
|
# [Q, K, V]
|
||
|
# to
|
||
|
# [Q1, Q2, K1, K2, V1, V2]
|
||
1 year ago
|
if is_transposed:
|
||
|
weight_chunks = torch.chunk(qkv, world_size * n_fused, dim=-1)
|
||
|
else:
|
||
|
weight_chunks = torch.chunk(qkv, world_size * n_fused, dim=0)
|
||
1 year ago
|
|
||
|
# rearrange the slice into the final order
|
||
|
# from
|
||
|
# [Q1, Q2, K1, K2, V1, V2]
|
||
|
# to
|
||
|
# [Q1, K1, V1], [Q2, K2, V2]
|
||
|
weight_chunks_of_current_rank = [weight_chunks[i] for i in order[rank::world_size]]
|
||
1 year ago
|
|
||
|
if is_transposed:
|
||
|
weight_of_current_rank = torch.cat(weight_chunks_of_current_rank, dim=-1)
|
||
|
else:
|
||
|
weight_of_current_rank = torch.cat(weight_chunks_of_current_rank, dim=0)
|
||
1 year ago
|
return weight_of_current_rank
|
||
|
|
||
|
|
||
1 year ago
|
def gather_fused_qkv_in_gpt2_style(qkv: torch.Tensor,
|
||
|
n_fused: int,
|
||
|
process_group: ProcessGroup,
|
||
|
is_transposed: bool = False):
|
||
1 year ago
|
"""
|
||
|
The splitted qkv tensor looks like [Q1, K1, V1] and [Q2, K2, V2], this function will gather them into [Q1, Q2, K1, K2, V1, V2].
|
||
1 year ago
|
|
||
|
Args:
|
||
|
qkv (torch.Tensor): The fused qkv tensor.
|
||
|
n_fused (int): The number items fused together, defaults to 3 (query, key and value).
|
||
|
process_group (ProcessGroup): The process group for distributed communication.
|
||
|
is_transposed (bool): generally the tensor is the shape of (out_features, in_features). Set this to True if the tensor is in the shape (in_features, out_features).
|
||
1 year ago
|
"""
|
||
|
world_size = dist.get_world_size(group=process_group)
|
||
|
|
||
|
# gather the tensors
|
||
|
# from
|
||
|
# [Q1, K1, V1], [Q2, K2, V2]
|
||
|
# to
|
||
|
# [Q1, K1, V1, Q2, K2, V2]
|
||
|
origin_device = qkv.device
|
||
|
qkv = qkv.cuda()
|
||
|
gather_list = [torch.zeros_like(qkv) for _ in range(world_size)]
|
||
|
dist.all_gather(gather_list, qkv, group=process_group)
|
||
1 year ago
|
|
||
|
if is_transposed:
|
||
|
gather_weight = torch.cat(gather_list, dim=-1)
|
||
|
else:
|
||
|
gather_weight = torch.cat(gather_list, dim=0)
|
||
1 year ago
|
gather_weight = gather_weight.to(origin_device)
|
||
|
qkv = qkv.to(origin_device)
|
||
|
|
||
|
# rearrange the tensor slices
|
||
|
# from
|
||
|
# [Q1, K1, V1, Q2, K2, V2]
|
||
|
# to
|
||
|
# [Q1, Q2, K1, K2, V1, V2]
|
||
1 year ago
|
if is_transposed:
|
||
|
weight_chunks = torch.chunk(gather_weight, world_size * n_fused, dim=-1)
|
||
|
else:
|
||
|
weight_chunks = torch.chunk(gather_weight, world_size * n_fused, dim=0)
|
||
|
|
||
1 year ago
|
reordered_chunk_list = []
|
||
|
for i in range(n_fused):
|
||
|
reordered_chunk_list.extend(weight_chunks[i::n_fused])
|
||
1 year ago
|
|
||
|
if is_transposed:
|
||
|
reordered_gather_weight = torch.cat(reordered_chunk_list, dim=-1)
|
||
|
else:
|
||
|
reordered_gather_weight = torch.cat(reordered_chunk_list, dim=0)
|
||
1 year ago
|
return reordered_gather_weight
|
||
|
|
||
|
|
||
1 year ago
|
class GPT2FusedLinearConv1D_Col(ParallelModule):
|
||
1 year ago
|
r"""Linear layer with column parallelism.
|
||
|
|
||
|
The linear layer is defined as :math:`Y = XA + b`. A is parallelized along
|
||
1 year ago
|
its second dimension as :math:`A = [A_1, ..., A_p]`. This layer is used to fit `Conv1D` layer (Fused QKV) in gpt2 of huggingface.
|
||
1 year ago
|
|
||
|
Args:
|
||
|
in_features (int): size of each input sample.
|
||
|
out_features (int): size of each output sample.
|
||
|
bias (bool, optional): If set to ``False``, the layer will not learn an additive bias, defaults to ``True``.
|
||
|
dtype (`torch.dtype`): The dtype of parameters, defaults to None.
|
||
|
device (`torch.device`): The device of parameters, defaults to None.
|
||
1 year ago
|
n_fused (int): The number items fused, defaults to 3 (QKV).
|
||
1 year ago
|
process_group (`torch.distributed.ProcessGroup`): The process group to be used for weight sharding and communication, defaults to None.
|
||
|
gather_output (bool, optional): If true, call all-gather on output and make Y available
|
||
|
to all GPUs, otherwise, every GPU will have its output
|
||
|
which is :math:`Y_i = XA_i`, defaults to False
|
||
|
skip_bias_add (bool): If set to ``True``, it will skip bias add for linear layer,
|
||
|
which is preserved for kernel fusion, defaults to False
|
||
|
weight_initializer (`typing.Callable`):
|
||
|
The initializer of weight, defaults to kaiming uniform initializer.
|
||
|
bias_initializer (`typing.Callable`):
|
||
|
The initializer of bias, defaults to xavier uniform initializer.
|
||
|
|
||
|
More details about ``initializer`` please refer to
|
||
|
`init <https://github.com/hpcaitech/ColossalAI/blob/main/colossalai/nn/init.py>`_.
|
||
|
"""
|
||
|
|
||
|
def __init__(self,
|
||
|
in_features: int,
|
||
|
out_features: int,
|
||
|
bias: bool = True,
|
||
|
dtype: torch.dtype = None,
|
||
|
device: torch.device = None,
|
||
|
process_group: ProcessGroup = None,
|
||
1 year ago
|
async_communication: bool = False,
|
||
1 year ago
|
gather_output: bool = False,
|
||
|
skip_bias_add: bool = False,
|
||
1 year ago
|
n_fused: int = 3,
|
||
1 year ago
|
weight_initializer: Callable = init.kaiming_uniform_(a=math.sqrt(5)),
|
||
|
bias_initializer: Callable = init.xavier_uniform_(a=1, scale=1)):
|
||
|
super().__init__()
|
||
|
|
||
|
# Keep input parameters
|
||
|
self.in_features = in_features
|
||
|
self.out_features = out_features
|
||
|
self.gather_output = gather_output
|
||
|
self.skip_bias_add = skip_bias_add
|
||
|
self.device = device
|
||
1 year ago
|
self.n_fused = n_fused
|
||
1 year ago
|
self.process_group = process_group
|
||
1 year ago
|
self.async_communication = async_communication
|
||
1 year ago
|
|
||
|
if skip_bias_add and not bias:
|
||
|
raise ValueError('cannot skip bias addition if bias is None')
|
||
|
|
||
|
# Parameters.
|
||
|
# Initialize weight.
|
||
|
factory_kwargs = {'device': device, 'dtype': dtype}
|
||
1 year ago
|
weight = torch.empty(self.in_features, self.out_features, **factory_kwargs)
|
||
|
|
||
|
def shard_fn(tensor):
|
||
1 year ago
|
return split_fused_qkv_in_gpt2_style(tensor, self.n_fused, self.process_group, True)
|
||
1 year ago
|
|
||
|
def gather_fn(tensor):
|
||
1 year ago
|
return gather_fused_qkv_in_gpt2_style(tensor, 3, self.process_group, True)
|
||
1 year ago
|
|
||
|
with torch.no_grad():
|
||
|
sharded_weight = distribute_tensor_with_customization(weight, shard_fn, gather_fn)
|
||
|
self.weight = customized_distributed_tensor_to_param(sharded_weight)
|
||
1 year ago
|
|
||
|
if bias:
|
||
1 year ago
|
bias = torch.empty(self.out_features, **factory_kwargs)
|
||
|
|
||
|
with torch.no_grad():
|
||
|
sharded_bias = distribute_tensor_with_customization(bias, shard_fn, gather_fn)
|
||
|
self.bias = customized_distributed_tensor_to_param(sharded_bias)
|
||
1 year ago
|
else:
|
||
|
self.bias = None
|
||
|
|
||
|
# offset the seed with randomizer index and rank
|
||
|
seed = torch.random.initial_seed()
|
||
|
self.randomizer = create_randomizer_with_offset(seed, process_group=self.process_group)
|
||
|
|
||
1 year ago
|
# init weights
|
||
|
self.reset_parameters(weight_initializer, bias_initializer)
|
||
1 year ago
|
|
||
|
@staticmethod
|
||
1 year ago
|
def from_native_module(module: nn.Module, process_group: Union[ProcessGroup, List[ProcessGroup]], n_fused: int,
|
||
1 year ago
|
*args, **kwargs) -> ParallelModule:
|
||
|
r"""
|
||
|
Convert a huggingface layer `Conv1D` in gpt2 to a parallelized linear layer.
|
||
1 year ago
|
|
||
|
Args:
|
||
|
module (`nn.Linear`): The module to be converted.
|
||
|
process_group (`Union[ProcessGroup, List[ProcessGroup]]`): The process group to be used for weight sharding and communication.
|
||
|
n_fused (int): The number of layers to be fused. In GPT2, Q,K,V are fused in one weight.
|
||
1 year ago
|
"""
|
||
|
# get the attributes
|
||
|
in_features = module.weight.shape[0]
|
||
|
out_features = module.weight.shape[1]
|
||
|
bias = module.bias is not None
|
||
|
device = module.weight.device
|
||
|
|
||
|
# ensure only one process group is passed
|
||
|
if isinstance(process_group, (list, tuple)):
|
||
|
assert len(process_group) == 1, \
|
||
|
f'Expected only one process group, got {len(process_group)}.'
|
||
|
process_group = process_group[0]
|
||
|
|
||
1 year ago
|
linear_1d = GPT2FusedLinearConv1D_Col(in_features=in_features,
|
||
|
out_features=out_features,
|
||
|
bias=bias,
|
||
|
device=device,
|
||
|
process_group=process_group,
|
||
|
*args,
|
||
|
**kwargs)
|
||
1 year ago
|
|
||
|
# TODO: copy the sharded weights
|
||
|
with torch.no_grad():
|
||
1 year ago
|
sharded_weight = split_fused_qkv_in_gpt2_style(module.weight.data,
|
||
|
n_fused=n_fused,
|
||
|
process_group=process_group,
|
||
|
is_transposed=True)
|
||
1 year ago
|
linear_1d.weight.data.copy_(sharded_weight.data)
|
||
1 year ago
|
|
||
|
if bias:
|
||
1 year ago
|
sharded_bias = split_fused_qkv_in_gpt2_style(module.bias.data,
|
||
|
n_fused=n_fused,
|
||
|
process_group=process_group,
|
||
|
is_transposed=True)
|
||
1 year ago
|
linear_1d.bias.data.copy_(sharded_bias.data)
|
||
1 year ago
|
|
||
|
return linear_1d
|
||
|
|
||
|
def reset_parameters(self, weight_initializer, bias_initializer) -> None:
|
||
1 year ago
|
with self.randomizer.fork_rng(enable_cpu=True):
|
||
|
fan_in, fan_out = self.in_features, self.out_features
|
||
|
weight_initializer(self.weight, fan_in=fan_in, fan_out=fan_out)
|
||
|
if self.bias is not None:
|
||
|
bias_initializer(self.bias, fan_in=fan_in)
|
||
1 year ago
|
|
||
|
def forward(self, input_: Tensor) -> Tuple[Tensor, Tensor]:
|
||
1 year ago
|
assert input_.shape[-1] == self.weight.shape[0], \
|
||
1 year ago
|
'Invalid shapes in Linear1D_Col forward: input={}, weight={}. Expected last dim of input {}.'.format(
|
||
|
input_.shape, self.weight.shape, self.weight.shape[-1])
|
||
|
# Set up backprop all-reduce.
|
||
1 year ago
|
input_parallel = reduce_backward(input_, self.process_group)
|
||
|
# input_parallel = input_
|
||
|
|
||
1 year ago
|
# Matrix multiply.
|
||
|
bias = self.bias if not self.skip_bias_add else None
|
||
1 year ago
|
|
||
|
output_parallel = matmul_with_async_comm(input_parallel, self.weight, bias, self.process_group,
|
||
|
self.async_communication)
|
||
1 year ago
|
|
||
|
if self.gather_output:
|
||
|
# All-gather across the partitions.
|
||
|
output = gather_forward_split_backward(output_parallel, dim=-1, process_group=self.process_group)
|
||
|
else:
|
||
|
output = output_parallel
|
||
|
|
||
|
if self.skip_bias_add:
|
||
|
return output, self.bias
|
||
|
else:
|
||
|
return output
|
||
|
|
||
|
|
||
1 year ago
|
class GPT2FusedLinearConv1D_Row(ParallelModule):
|
||
1 year ago
|
r""" Linear layer with row parallelism.
|
||
|
This layer is used to fit `Conv1D` layer (Fused QKV) in gpt2 of huggingface.
|
||
1 year ago
|
|
||
|
Args:
|
||
|
in_features (int): size of each input sample.
|
||
|
out_features (int): size of each output sample.
|
||
|
bias (bool, optional): If set to ``False``, the layer will not learn an additive bias, defaults to ``True``.
|
||
|
dtype (`torch.dtype`): The dtype of parameters, defaults to None.
|
||
|
parallel_input (bool): If set to ``True``, it's assumed that the input is split, defaults to False.
|
||
|
skip_bias_add (bool): If set to ``True``, it will skip bias add for linear layer,
|
||
|
which is preserved for kernel fusion, defaults to False
|
||
|
weight_initializer (:class:`typing.Callable`, optional):
|
||
|
The initializer of weight, defaults to kaiming uniform initializer.
|
||
|
bias_initializer (:class:`typing.Callable`, optional):
|
||
|
The initializer of bias, defaults to xavier uniform initializer.
|
||
|
|
||
|
More details about ``initializer`` please refer to
|
||
|
`init <https://github.com/hpcaitech/ColossalAI/blob/main/colossalai/nn/init.py>`_.
|
||
|
"""
|
||
|
|
||
|
def __init__(self,
|
||
|
in_features: int,
|
||
|
out_features: int,
|
||
|
bias: bool = True,
|
||
|
dtype: torch.dtype = None,
|
||
|
device: torch.device = None,
|
||
|
process_group: ProcessGroup = None,
|
||
|
parallel_input: bool = True,
|
||
|
skip_bias_add: bool = False,
|
||
|
weight_initializer: Callable = init.kaiming_uniform_(a=math.sqrt(5)),
|
||
|
bias_initializer: Callable = init.xavier_uniform_(a=1, scale=1),
|
||
|
stream_chunk_num: int = 1):
|
||
|
super().__init__()
|
||
|
|
||
|
self.stream_chunk_num = stream_chunk_num
|
||
|
|
||
|
# Keep input parameters
|
||
|
self.in_features = in_features
|
||
|
self.out_features = out_features
|
||
|
self.parallel_input = parallel_input
|
||
|
self.skip_bias_add = skip_bias_add
|
||
|
self.process_group = process_group
|
||
|
self.num_partitions = dist.get_world_size(self.process_group)
|
||
|
|
||
|
if skip_bias_add and not bias:
|
||
|
raise ValueError('cannot skip bias addition if bias is None')
|
||
|
|
||
|
# Divide the weight matrix along the last dimension.
|
||
|
self.input_size_per_partition = divide(in_features, self.num_partitions)
|
||
|
|
||
|
# Parameters.
|
||
|
# Initialize weight.
|
||
|
factory_kwargs = {'device': device, 'dtype': dtype}
|
||
1 year ago
|
weight = torch.empty(self.in_features, self.out_features, **factory_kwargs)
|
||
|
sharded_weight = shard_rowwise(weight, self.process_group)
|
||
|
self.weight = sharded_tensor_to_param(sharded_weight)
|
||
1 year ago
|
|
||
|
if self.stream_chunk_num > 1:
|
||
|
# TODO() work for inference only
|
||
|
self.chunk_weight()
|
||
|
if bias:
|
||
|
self.bias = Parameter(torch.empty(self.out_features, **factory_kwargs))
|
||
|
else:
|
||
|
self.bias = None
|
||
|
|
||
|
# offset the seed with randomizer index and rank
|
||
|
seed = torch.random.initial_seed()
|
||
|
self.randomizer = create_randomizer_with_offset(seed, process_group=self.process_group)
|
||
|
|
||
1 year ago
|
# init weights
|
||
|
self.reset_parameters(weight_initializer, bias_initializer)
|
||
1 year ago
|
|
||
|
@staticmethod
|
||
1 year ago
|
def from_native_module(module: nn.Linear, process_group: Union[ProcessGroup, List[ProcessGroup]], *args,
|
||
|
**kwargs) -> ParallelModule:
|
||
1 year ago
|
r"""
|
||
|
Convert a native PyTorch linear layer to a parallelized linear layer.
|
||
|
"""
|
||
|
# get the attributes
|
||
|
in_features = module.weight.shape[0]
|
||
|
out_features = module.weight.shape[1]
|
||
|
bias = module.bias is not None
|
||
|
device = module.weight.device
|
||
|
|
||
|
# ensure only one process group is passed
|
||
|
if isinstance(process_group, (list, tuple)):
|
||
|
assert len(process_group) == 1, \
|
||
|
f'Expected only one process group, got {len(process_group)}.'
|
||
|
process_group = process_group[0]
|
||
|
|
||
1 year ago
|
linear_1d = GPT2FusedLinearConv1D_Row(in_features=in_features,
|
||
|
out_features=out_features,
|
||
|
bias=bias,
|
||
|
device=device,
|
||
|
process_group=process_group,
|
||
|
*args,
|
||
|
**kwargs)
|
||
1 year ago
|
|
||
|
# TODO: copy the sharded weights
|
||
|
with torch.no_grad():
|
||
|
# the weigh to the linear layer is a transpose
|
||
|
# thus shard on col is equal to shard on row
|
||
1 year ago
|
sharded_weight = shard_rowwise(module.weight.data, process_group)
|
||
1 year ago
|
linear_1d.weight.data.copy_(sharded_weight.data)
|
||
1 year ago
|
|
||
|
if bias:
|
||
1 year ago
|
linear_1d.bias.copy_(module.bias.data)
|
||
1 year ago
|
|
||
|
return linear_1d
|
||
|
|
||
|
def chunk_weight(self):
|
||
|
self.weight_list = torch.chunk(self.weight, self.stream_chunk_num, dim=0)
|
||
|
|
||
|
def reset_parameters(self, weight_initializer, bias_initializer) -> None:
|
||
1 year ago
|
with self.randomizer.fork_rng(enable_cpu=True):
|
||
|
fan_in, fan_out = self.in_features, self.out_features
|
||
|
weight_initializer(self.weight, fan_in=fan_in, fan_out=fan_out)
|
||
|
|
||
|
if self.bias is not None:
|
||
|
bias_initializer(self.bias, fan_in=fan_in)
|
||
|
if self.process_group is None:
|
||
|
src_rank = 0
|
||
|
else:
|
||
|
src_rank = dist.distributed_c10d._get_global_rank(self.process_group, 0)
|
||
|
|
||
|
origin_device = self.bias.device
|
||
|
self.bias = self.bias.cuda()
|
||
|
dist.broadcast(self.bias, src=src_rank, group=self.process_group)
|
||
|
self.bias = self.bias.to(origin_device)
|
||
1 year ago
|
|
||
|
def forward(self, input_: Tensor) -> Tensor:
|
||
|
# Set up backprop all-reduce.
|
||
|
if self.parallel_input:
|
||
1 year ago
|
assert input_.shape[-1] == self.weight.shape[0], \
|
||
1 year ago
|
'Invalid shapes in Linear1D_Row forward: input={}, weight={}. Expected last dim of input {}.'.format(
|
||
|
input_.shape, self.weight.shape, self.weight.shape[-1])
|
||
|
input_ = input_
|
||
|
else:
|
||
1 year ago
|
assert divide(input_.shape[-1], self.num_partitions) == self.weight.shape[0], \
|
||
1 year ago
|
'Invalid shapes in Linear1D_Row forward: input={}, weight={}. Expected last dim of input {}.'.format(
|
||
|
input_.shape, self.weight.shape, self.weight.shape[-1] * self.num_partitions)
|
||
|
input_ = split_forward_gather_backward(input_, dim=-1, process_group=self.process_group)
|
||
|
|
||
|
if self.stream_chunk_num > 1:
|
||
|
if self.training:
|
||
|
raise RuntimeError("use stream_chunk_num=1 in Linear1D_Row for training!")
|
||
|
with torch.no_grad():
|
||
|
output_parallel_list = [None for i in range(self.stream_chunk_num)]
|
||
|
handle_list = []
|
||
|
for i in range(self.stream_chunk_num):
|
||
1 year ago
|
output_parallel_list[i] = torch.matmul(input_, self.weight_list[i])
|
||
1 year ago
|
handle = torch.distributed.all_reduce(output_parallel_list[i],
|
||
|
group=self.process_group,
|
||
|
async_op=True)
|
||
|
handle_list.append(handle)
|
||
|
# output_parallel_list[i] = reduce_input(output_parallel_list[i], ParallelMode.PARALLEL_1D)
|
||
|
for handle in handle_list:
|
||
|
handle.wait()
|
||
|
output = torch.cat(output_parallel_list, dim=-1)
|
||
|
else:
|
||
1 year ago
|
output_parallel = torch.matmul(input_, self.weight)
|
||
|
output = reduce_forward(output_parallel, self.process_group)
|
||
1 year ago
|
|
||
|
if not self.skip_bias_add:
|
||
|
if self.bias is not None:
|
||
|
output = output + self.bias
|
||
|
return output
|
||
|
else:
|
||
|
return output, self.bias
|