Skip to content

Offloading optimizer to CPU causes "expected input to be on cuda" error; Suggest to fallback to torch.optim.AdamW #71

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
hibagus wants to merge 1 commit into from
Closed

Offloading optimizer to CPU causes "expected input to be on cuda" error; Suggest to fallback to torch.optim.AdamW #71

hibagus wants to merge 1 commit into from

Conversation

@hibagus
Copy link

@hibagus hibagus commented Aug 7, 2022

When enabling CPU Offloading for the Optimizer as follows:

offload_optimizer": {
      "offload_optimizer": {
            "device": "cpu",            # "[cpu|nvme]"
            "nvme_path": "/local_nvme",
            "pin_memory": false,         # [true|false]
            "buffer_count": 4,
            "fast_init": false
        },

the following error occurs:

Traceback (most recent call last):
  File "/home/bagus/DeepSpeed/Debug/Megatron-DeepSpeed/pretrain_gpt.py", line 276, in <module>
    pretrain(train_valid_test_datasets_provider, model_provider, forward_step,
  File "/home/bagus/DeepSpeed/Debug/Megatron-DeepSpeed/megatron/training.py", line 130, in pretrain
    model, optimizer, lr_scheduler = setup_model_and_optimizer(model_provider, teacher=False)
  File "/home/bagus/DeepSpeed/Debug/Megatron-DeepSpeed/megatron/training.py", line 420, in setup_model_and_optimizer
    model, optimizer, _, lr_scheduler = deepspeed.initialize(
  File "/home/bagus/anaconda3/envs/MegatronDeepSpeed/lib/python3.8/site-packages/deepspeed/__init__.py", line 121, in initialize
    engine = DeepSpeedEngine(args=args,
  File "/home/bagus/anaconda3/envs/MegatronDeepSpeed/lib/python3.8/site-packages/deepspeed/runtime/engine.py", line 310, in __init__
    self._configure_optimizer(optimizer, model_parameters)
  File "/home/bagus/anaconda3/envs/MegatronDeepSpeed/lib/python3.8/site-packages/deepspeed/runtime/engine.py", line 1096, in _configure_optimizer
    self.optimizer = self._configure_zero_optimizer(basic_optimizer)
  File "/home/bagus/anaconda3/envs/MegatronDeepSpeed/lib/python3.8/site-packages/deepspeed/runtime/engine.py", line 1348, in _configure_zero_optimizer
    optimizer = DeepSpeedZeroOptimizer(
  File "/home/bagus/anaconda3/envs/MegatronDeepSpeed/lib/python3.8/site-packages/deepspeed/runtime/zero/stage_1_and_2.py", line 510, in __init__
    self.initialize_optimizer_states()
  File "/home/bagus/anaconda3/envs/MegatronDeepSpeed/lib/python3.8/site-packages/deepspeed/runtime/zero/stage_1_and_2.py", line 600, in initialize_optimizer_states
    self.optimizer.step()
  File "/home/bagus/anaconda3/envs/MegatronDeepSpeed/lib/python3.8/site-packages/torch/optim/optimizer.py", line 109, in wrapper
    return func(*args, **kwargs)
  File "/home/bagus/anaconda3/envs/MegatronDeepSpeed/lib/python3.8/site-packages/apex/optimizers/fused_adam.py", line 180, in step
    multi_tensor_applier(self.multi_tensor_adam,
  File "/home/bagus/anaconda3/envs/MegatronDeepSpeed/lib/python3.8/site-packages/apex/multi_tensor_apply/multi_tensor_apply.py", line 27, in __call__
    return op(self.chunk_size,
RuntimeError: expected input to be on cuda

It seems like args.cpu_optimizer is always set to False, although the CPU optimizer offloading is enabled. Therefore, the FusedAdam optimizer from apex.optimizer is always used. This optimizer expects input on GPU memory, not on CPU memory.

As a temporary fix, I would suggest to fallback to torch.optim.AdamW which can be used both on CPU and GPU. At least, until we can find out why args.cpu_optimizer is always set to False.

Let me know what you think :)

My environment:
OS: Ubuntu 20.04
GPU: 4xA100 40GB SXM4
ds_report:

--------------------------------------------------
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
--------------------------------------------------
cpu_adam ............... [NO] ....... [OKAY]
cpu_adagrad ............ [NO] ....... [OKAY]
fused_adam ............. [NO] ....... [OKAY]
fused_lamb ............. [NO] ....... [OKAY]
sparse_attn ............ [NO] ....... [OKAY]
transformer ............ [NO] ....... [OKAY]
stochastic_transformer . [NO] ....... [OKAY]
async_io ............... [NO] ....... [OKAY]
utils .................. [NO] ....... [OKAY]
quantizer .............. [NO] ....... [OKAY]
transformer_inference .. [NO] ....... [OKAY]
--------------------------------------------------
DeepSpeed general environment info:
torch install path ............... ['/home/bagus/anaconda3/envs/MegatronDeepSpeed/lib/python3.8/site-packages/torch']
torch version .................... 1.12.0+cu113
torch cuda version ............... 11.3
torch hip version ................ None
nvcc version ..................... 11.6
deepspeed install path ........... ['/home/bagus/anaconda3/envs/MegatronDeepSpeed/lib/python3.8/site-packages/deepspeed']
deepspeed info ................... 0.6.7, unknown, unknown
deepspeed wheel compiled w. ...... torch 1.12, cuda 11.3

@hibagus
use torch.optim.AdamW instead of FusedAdam from apex.optimizer as a t…
8fd352d 
…emporary solution for CPU optimizer offloading
@ghost
Copy link

ghost commented Aug 7, 2022 •
edited by ghost
Loading

CLA assistant check
All CLA requirements met.

@hibagus hibagus changed the title Offloading optimizer to CPU causes expected input to be on cuda; Suggest to fallback to torch.optim.AdamW Offloading optimizer to CPU causes "expected input to be on cuda" error; Suggest to fallback to torch.optim.AdamW Aug 7, 2022
@hibagus hibagus mentioned this pull request Aug 8, 2022
Closed
@tjruwase tjruwase mentioned this pull request Aug 10, 2022
Merged
@awan-10
Copy link

awan-10 commented Aug 12, 2022

@hibagus - this has been fixed in the latest deepspeed. Can you kindly test and close this PR?

@hibagus
Copy link
Author

hibagus commented Aug 13, 2022

@awan-10 Yes that fix solve the problem. I will close this PR. Thanks Awan!

@hibagus hibagus closed this Aug 13, 2022
saforem2 added a commit to saforem2/Megatron-DeepSpeed that referenced this pull request Dec 25, 2024
@saforem2
Merge pull request deepspeedai#71 from argonne-lcf/aurora-post-at
587aafd
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@jeffra jeffra Awaiting requested review from jeffra jeffra is a code owner

@samyam samyam Awaiting requested review from samyam

@tjruwase tjruwase Awaiting requested review from tjruwase tjruwase is a code owner

@ShadenSmith ShadenSmith Awaiting requested review from ShadenSmith

@conglongli conglongli Awaiting requested review from conglongli

@awan-10 awan-10 Awaiting requested review from awan-10

@cli99 cli99 Awaiting requested review from cli99

@eltonzheng eltonzheng Awaiting requested review from eltonzheng

@minjiaz minjiaz Awaiting requested review from minjiaz

@RezaYazdaniAminabadi RezaYazdaniAminabadi Awaiting requested review from RezaYazdaniAminabadi

@duli2012 duli2012 Awaiting requested review from duli2012

@mrwyattii mrwyattii Awaiting requested review from mrwyattii

@yaozhewei yaozhewei Awaiting requested review from yaozhewei

@arashb arashb Awaiting requested review from arashb

@xiaoxiawu-microsoft xiaoxiawu-microsoft Awaiting requested review from xiaoxiawu-microsoft

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@hibagus @awan-10