inkcherry c66bc4269e set the default to use set_to_none for clearing gradients in BF16 optimizer. (#5434) 6 月之前
..
activation_checkpointing 08e0733e4a Support MoE for pipeline models (#5338) 6 月之前
checkpoint_engine c5edc91ecb change partititon_name to partition_name (#3700) 1 年之前
comm 6dcced1d5c Cleanup required_torch_version code and references. (#5370) 6 月之前
compression b361c72761 Update DeepSpeed copyright license to Apache 2.0 (#3111) 1 年之前
data_pipeline 64defe65b7 Parallel map step for `DistributedDataAnalyzer` map-reduce (#5291) 6 月之前
fp16 0896503e2f Fix a convergence issues in TP topology caused by incorrect grad_norm. (#5411) 6 月之前
pipe 6dcced1d5c Cleanup required_torch_version code and references. (#5370) 6 月之前
swap_tensor d058d4b39b Nvme offload checkpoint (#4707) 9 月之前
zero 54c0687264 stage3: efficient compute of scaled_global_grad_norm (#5256) 6 月之前
__init__.py c56a4b9e0d Improve universal checkpoint (#5289) 6 月之前
base_optimizer.py c56a4b9e0d Improve universal checkpoint (#5289) 6 月之前
bf16_optimizer.py c66bc4269e set the default to use set_to_none for clearing gradients in BF16 optimizer. (#5434) 6 月之前
compiler.py d274ebf347 allow debug/experimental compiler backends (#5191) 7 月之前
config.py ed8aed5703 fix comms dtype (#5297) 6 月之前
config_utils.py 604d701e35 Introduce pydantic_v1 compatibility module for pydantic>=2.0.0 support (#4407) 1 年之前
constants.py 3c0bd31288 BF16 optimizer: Improve device utilization by immediate grad update (#4975) 8 月之前
dataloader.py b361c72761 Update DeepSpeed copyright license to Apache 2.0 (#3111) 1 年之前
eigenvalue.py b361c72761 Update DeepSpeed copyright license to Apache 2.0 (#3111) 1 年之前
engine.py bc0f774728 Update engine.py to avoid torch warning (#5408) 6 月之前
hybrid_engine.py 5f41bd06dd Fix Hybrid Engine metrics printing (#4789) 10 月之前
lr_schedules.py ce0ebdade2 [Bug fix] WarmupCosineLR issues (#4688) 11 月之前
progressive_layer_drop.py b361c72761 Update DeepSpeed copyright license to Apache 2.0 (#3111) 1 年之前
quantize.py b361c72761 Update DeepSpeed copyright license to Apache 2.0 (#3111) 1 年之前
sparse_tensor.py c84c28d23b Support cpu tensors without direct device invocation (#3842) 9 月之前
state_dict_factory.py b361c72761 Update DeepSpeed copyright license to Apache 2.0 (#3111) 1 年之前
utils.py 0896503e2f Fix a convergence issues in TP topology caused by incorrect grad_norm. (#5411) 6 月之前
weight_quantizer.py b361c72761 Update DeepSpeed copyright license to Apache 2.0 (#3111) 1 年之前