[model checkpoint] updated checkpoint save/load utils (#592)

This commit is contained in:
アマデウス
2022-04-01 16:49:21 +08:00
committed by GitHub
parent 1c40ee8749
commit acae68eb04
2 changed files with 218 additions and 176 deletions

View File

@@ -1,9 +1,9 @@
from .cuda import empty_cache, get_current_device, set_to_cuda, synchronize
from .activation_checkpoint import checkpoint
from .checkpointing import load_checkpoint, save_checkpoint
from .common import (clip_grad_norm_fp32, conditional_context, copy_tensor_parallel_attributes, count_zeros_fp32,
free_port, is_dp_rank_0, is_model_parallel_parameter, is_no_pp_or_last_stage, is_tp_rank_0,
is_using_ddp, is_using_pp, is_using_sequence, multi_tensor_applier,
ensure_path_exists, free_port, is_dp_rank_0, is_model_parallel_parameter, is_no_pp_or_last_stage,
is_tp_rank_0, is_using_ddp, is_using_pp, is_using_sequence, multi_tensor_applier,
param_is_not_tensor_parallel_duplicate, print_rank_0, switch_virtual_pipeline_parallel_rank,
sync_model_param)
from .data_sampler import DataParallelSampler, get_dataloader
@@ -18,5 +18,6 @@ __all__ = [
'is_model_parallel_parameter', 'clip_grad_norm_fp32', 'count_zeros_fp32', 'copy_tensor_parallel_attributes',
'param_is_not_tensor_parallel_duplicate', 'get_current_device', 'synchronize', 'empty_cache', 'set_to_cuda',
'report_memory_usage', 'Timer', 'MultiTimer', 'multi_tensor_applier', 'accumulate_gradient', 'DataParallelSampler',
'get_dataloader', 'switch_virtual_pipeline_parallel_rank', 'TensorDetector'
'get_dataloader', 'switch_virtual_pipeline_parallel_rank', 'TensorDetector', 'load_checkpoint', 'save_checkpoint',
'ensure_path_exists'
]