Index

_

__del__() (fairscale.optim.AdaScale method) __getattr__() (fairscale.nn.FullyShardedDataParallel method) (fairscale.nn.ShardedDataParallel method) (fairscale.optim.AdaScale method) __getitem__() (fairscale.nn.FullyShardedDataParallel method) (fairscale.nn.Pipe method)

__getstate__() (fairscale.nn.FullyShardedDataParallel method) __iter__() (fairscale.nn.Pipe method) __len__() (fairscale.nn.Pipe method) __setstate__() (fairscale.nn.FullyShardedDataParallel method)

A

AdaScale (class in fairscale.optim) add_param_group() (fairscale.optim.AdaScale method) (fairscale.optim.OSS method)

append_shared_param() (fairscale.nn.FullyShardedDataParallel method) apply() (fairscale.nn.FullyShardedDataParallel method) assert_state() (fairscale.nn.FullyShardedDataParallel method)

B

balance (fairscale.nn.Pipe attribute)

C

checkpoint (fairscale.nn.Pipe attribute) checkpoint_wrapper (class in fairscale.nn.checkpoint) chunks (fairscale.nn.Pipe attribute) clip_grad_norm() (fairscale.optim.OSS method) clip_grad_norm_() (fairscale.nn.FullyShardedDataParallel method)

consolidate_shard_weights() (fairscale.nn.FullyShardedDataParallel static method) consolidate_state_dict() (fairscale.optim.OSS method) cpu() (fairscale.nn.Pipe method) cpu_offload (fairscale.nn.FullyShardedDataParallel property) cuda() (fairscale.nn.Pipe method)

D

devices (fairscale.nn.Pipe attribute)

E

extra_repr() (fairscale.nn.FullyShardedDataParallel method)

F

forward() (fairscale.experimental.nn.OffloadModel method) (fairscale.nn.FullyShardedDataParallel method) (fairscale.nn.Pipe method) (fairscale.nn.ShardedDataParallel method)

FullyShardedDataParallel (class in fairscale.nn)

G

gain() (fairscale.optim.AdaScale method)

gather_full_optim_state_dict() (fairscale.nn.FullyShardedDataParallel method) get_shard_from_optim_state_dict() (fairscale.nn.FullyShardedDataParallel method)

I

in_super_constructor (fairscale.optim.OSS attribute)

L

load_local_state_dict() (fairscale.nn.FullyShardedDataParallel method) load_state_dict() (fairscale.nn.FullyShardedDataParallel method) (fairscale.optim.AdaScale method) (fairscale.optim.OSS method)

local_metadata_dict() (fairscale.nn.FullyShardedDataParallel method) local_state_dict() (fairscale.nn.FullyShardedDataParallel method)

M

module (fairscale.nn.FullyShardedDataParallel property)

MOELayer (class in fairscale.nn)

N

named_parameters() (fairscale.nn.FullyShardedDataParallel method) no_sync() (fairscale.nn.FullyShardedDataParallel method) (fairscale.nn.ShardedDataParallel method)

non_shared_params() (fairscale.nn.FullyShardedDataParallel method)

O

OffloadModel (class in fairscale.experimental.nn)

optim (fairscale.optim.OSS attribute) OSS (class in fairscale.optim)

P

parameters() (fairscale.nn.FullyShardedDataParallel method) params_with_grad (fairscale.nn.FullyShardedDataParallel property)

partition_parameters() (fairscale.optim.OSS method) perform_slowmo() (fairscale.experimental.nn.data_parallel.SlowMoDistributedDataParallel method) Pipe (class in fairscale.nn)

R

reduce() (fairscale.nn.ShardedDataParallel method) refresh_trainable() (fairscale.nn.ShardedDataParallel method) (fairscale.optim.OSS method)

reset_parameters() (fairscale.nn.FullyShardedDataParallel method)

S

scale (fairscale.optim.AdaScale property) scale_grad_by_num_grads_to_accum() (fairscale.optim.AdaScale method) set_gradient_divide_factors() (fairscale.nn.FullyShardedDataParallel method) set_num_gradients_to_accumulate() (fairscale.optim.AdaScale method) set_scale() (fairscale.optim.AdaScale method) ShardedDataParallel (class in fairscale.nn) SlowMoDistributedDataParallel (class in fairscale.experimental.nn.data_parallel)

smoothing (fairscale.optim.AdaScale property) state_dict() (fairscale.nn.FullyShardedDataParallel method) (fairscale.optim.AdaScale method) (fairscale.optim.OSS method) step() (fairscale.optim.AdaScale method) (fairscale.optim.OSS method) summon_full_params() (fairscale.nn.FullyShardedDataParallel method) sync_buffers() (fairscale.nn.ShardedDataParallel method)

T

to() (fairscale.nn.Pipe method) (fairscale.nn.ShardedDataParallel method)

training (fairscale.experimental.nn.OffloadModel attribute) (fairscale.nn.FullyShardedDataParallel attribute) (fairscale.nn.ShardedDataParallel attribute)

U

unhook() (fairscale.optim.AdaScale method)

Z

zero_grad() (fairscale.nn.ShardedDataParallel method) (fairscale.optim.AdaScale method)