[deepspeed] zero inference #14253
Merged
[deepspeed] zero inference #14253
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
5 tasks
sgugger
approved these changes
Nov 20, 2021
|
|
||
| Inference: | ||
|
|
||
| 1. DeepSpeed ZeRO Inference - same as Training but doesn't require Optimizer |
There was a problem hiding this comment.
Can we make a real sentence? I don't understand what this means.
| @@ -111,6 +111,24 @@ def get_value(self, ds_key_long, default=None): | |||
| return default | |||
| return config.get(ds_key, default) | |||
|
|
|||
| def del_config_sub_tree(self, ds_key_long, must_exist=False): | |||
| config = self.config | |||
There was a problem hiding this comment.
This method deservers a docstring.
src/transformers/deepspeed.py
Outdated
| config = config.get(node) | ||
| if config is None: | ||
| if must_exist: | ||
| raise ValueError(f"Can't find {ds_key_long} entry in the config") |
There was a problem hiding this comment.
Might be worth printing the config as well?
Suggested change
| raise ValueError(f"Can't find {ds_key_long} entry in the config") | |
| raise ValueError(f"Can't find {ds_key_long} entry in the config {config}") |
There was a problem hiding this comment.
that won't work, but I will fix it to dump the config, thank you.
| hf_deepspeed_config.trainer_config_finalize(args, model, num_training_steps) | ||
|
|
||
| # resume config update - some bits like `model` and `num_training_steps` only become available during train | ||
| def deepspeed_optim_sched(trainer, hf_deepspeed_config, args, num_training_steps): |
There was a problem hiding this comment.
This new method deserves a docstring.
Co-authored-by: Sylvain Gugger <[email protected]>
|
Thanks a lot for the review and the suggestions, Sylvain. All addressed, please have another look at your convenience. Thank you. |
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.
This PR extends HF/DS integration to support Deepspeed Zero inference. Now we don't need to waste gpu memory on allocating the optimizer/scheduler and then dropping them. And in some cases enabling what was not possible before - in case the user doesn't have the extra gpu memory and was getting OOM on inference.
Blocking events:
@jeffra, @sgugger
The CI errors seem to be irrelevant to this PR