Select the DeepSpeedCPUOptimizer based on the original optimizer class. #3255
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Thanks for this update. Just wondering aloud: If the optimizer is neither Adam, nor Adagrad, nor Lion, what would be the right thing to do? As is, Adam is used, is that correct? |
|
@BenjaminBossan By default, I have kept the Adam as it was for everyone torch optimizer. |
muellerzr
reviewed
Nov 25, 2024
src/accelerate/accelerator.py
Outdated
Comment on lines
1852
to
1879
| # For DeepSpeedCPUAdagrad | ||
| if compare_versions("deepspeed", ">=", "0.5.5"): | ||
| # Check if the optimizer is PyTorch's Adagrad. | ||
| is_ada = isinstance(optimizer, torch.optim.Adagrad) | ||
| # If not, and bitsandbytes is available, | ||
| # # check if the optimizer is the 32-bit bitsandbytes Adagrad. | ||
| if is_bnb_available() and not is_ada: | ||
| import bitsandbytes.optim as bnb_opt | ||
|
|
||
| is_ada = ( | ||
| isinstance(optimizer, (bnb_opt.Adagrad, bnb_opt.Adagrad32bit)) | ||
| and optimizer.optim_bits == 32 | ||
| ) | ||
| if is_ada: | ||
| from deepspeed.ops.adagrad import DeepSpeedCPUAdagrad | ||
|
|
||
| optimizer_class = DeepSpeedCPUAdagrad | ||
|
|
||
| # For DeepSpeedCPULion | ||
| if is_bnb_available(min_version="0.38.0") and compare_versions("deepspeed", ">=", "0.11.0"): | ||
| from bitsandbytes.optim import Lion, Lion32bit | ||
|
|
||
| if isinstance(optimizer, (Lion, Lion32bit)) and optimizer.optim_bits == 32: | ||
| from deepspeed.ops.lion import DeepSpeedCPULion | ||
|
|
||
| optimizer_class = DeepSpeedCPULion | ||
|
|
||
| optimizer = optimizer_class(optimizer.param_groups, **defaults) |
There was a problem hiding this comment.
This seems a bit bulky for the Accelerator, can we abstract this out to a deepspeed util of map_pytorch_optim_to_deepspeed?
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
|
@muellerzr What do you think of it like this? |
muellerzr
approved these changes
Dec 2, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
For DeepSpeed optimizer offloading, select the appropriate optimizer based on the original optimizer class.
Who can review?