[BUG] DeepSpeed GradBucket #7393
Description
Describe the bug
I am following the tutorial on communication logging and ran into an issue with replaces torch.distributed as dist with deepspeed.comm as dist. In the tutorial, it states that "(Optional) If your application contains torch.distributed calls that you wish to log, import deepspeed.comm package and modify torch.distributed calls to use deepspeed.comm (Note: The deepspeed.comm collective and pt2pt APIs exactly match torch.distributed)". We are using GradBucket at a type hint for reduction comm hook, and apparently this isn't implemented as part of the deepspeed.comm api, though it is for pytorch.
To Reproduce
Steps to reproduce the behavior:
import deepspeed.comm as dist
test = dist.GradBucket()Expected behavior
GradBucket should be exposed in the api
ds_report output
--------------------------------------------------
DeepSpeed C++/CUDA extension op report
--------------------------------------------------
NOTE: Ops not installed will be just-in-time (JIT) compiled at
runtime if needed. Op compatibility means that your system
meet the required dependencies to JIT install the op.
--------------------------------------------------
JIT compiled ops requires ninja
ninja .................. [OKAY]
--------------------------------------------------
op name ................ installed .. compatible
--------------------------------------------------
async_io ............... [NO] ....... [OKAY]
fused_adam ............. [NO] ....... [OKAY]
cpu_adam ............... [NO] ....... [OKAY]
cpu_adagrad ............ [NO] ....... [OKAY]
cpu_lion ............... [NO] ....... [OKAY]
dc ..................... [NO] ....... [OKAY]
[WARNING] Please specify the CUTLASS repo directory as environment variable $CUTLASS_PATH
evoformer_attn ......... [NO] ....... [NO]
[WARNING] FP Quantizer is using an untested triton version (3.3.1), only 2.3.(0, 1) and 3.0.0 are known to be compatible with these kernels
fp_quantizer ........... [NO] ....... [NO]
fused_lamb ............. [NO] ....... [OKAY]
fused_lion ............. [NO] ....... [OKAY]
gds .................... [NO] ....... [OKAY]
transformer_inference .. [NO] ....... [OKAY]
inference_core_ops ..... [NO] ....... [OKAY]
cutlass_ops ............ [NO] ....... [OKAY]
quantizer .............. [NO] ....... [OKAY]
ragged_device_ops ...... [NO] ....... [OKAY]
ragged_ops ............. [NO] ....... [OKAY]
random_ltd ............. [NO] ....... [OKAY]
[WARNING] sparse_attn requires a torch version >= 1.5 and < 2.0 but detected 2.7
[WARNING] using untested triton version (3.3.1), only 1.0.0 is known to be compatible
sparse_attn ............ [NO] ....... [NO]
spatial_inference ...... [NO] ....... [OKAY]
transformer ............ [NO] ....... [OKAY]
stochastic_transformer . [NO] ....... [OKAY]
utils .................. [NO] ....... [OKAY]
--------------------------------------------------
DeepSpeed general environment info:
torch install path ............... ['/home/alexk101/Documents/phd/projects/scaling-transformers-physical-sciences/.venv/lib/python3.11/site-packages/torch']
torch version .................... 2.7.1+cu126
deepspeed install path ........... ['/home/alexk101/Documents/phd/projects/scaling-transformers-physical-sciences/.venv/lib/python3.11/site-packages/deepspeed']
deepspeed info ................... 0.17.1, unknown, unknown
torch cuda version ............... 12.6
torch hip version ................ None
nvcc version ..................... 12.9
deepspeed wheel compiled w. ...... torch 0.0, cuda 0.0
shared memory (/dev/shm) size .... 45.90 GB
Screenshots
If applicable, add screenshots to help explain your problem.
System info (please complete the following information):
- OS: SUSE Linux Enterprise Server 15 SP6 x86_64
- GPU count and types: 8x MI250x * 64 nodes
- Interconnects (if applicable): HP Slingshot
- Python version: 3.11
Launcher context
slurm srun
Docker context
N/A
Additional context
Running on Frontier, though this the platform isn't really related to this issue, more a question of why the API here doesn't match pytorch