[amp] refine transformer fp16 train #1574
Merged
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
FrostML
reviewed
Jan 10, 2022
| scaler = paddle.amp.GradScaler( | ||
| init_loss_scaling=args.scale_loss) | ||
| with paddle.amp.auto_cast(): | ||
| with paddle.amp.auto_cast(custom_black_list={'scale', 'reduce_sum', 'elementwise_div'} if amp_level=='O2' else {}, level=amp_level): |
FrostML
reviewed
Jan 11, 2022
| @@ -71,6 +72,11 @@ def parse_args(): | |||
| default=None, | |||
| type=str, | |||
| help="The eos token. It should be provided when use custom vocab_file. ") | |||
| parser.add_argument( | |||
There was a problem hiding this comment.
若使用 arg parser 的方式新增 --num_workers，注意还需要在：
tests/transformer/train.pyexamples/machine_translation/transformer/predict.pyexamples/machine_translation/transformer/static/train.pyexamples/machine_translation/transformer/static/predict.pyexamples/machine_translation/transformer/deploy/python/inference.pyexamples/machine_translation/transformer/faster_transformer/encoder_decoding_predict.py
也需要补充下。
或是加在 yaml 文件：
examples/machine_translation/transformer/configs/transformer.base.yamlexamples/machine_translation/transformer/configs/transformer.big.yaml
28 行，Args for reader的位置。
This was referenced Jan 25, 2022
FrostML
added a commit
that referenced
this pull request
Jan 25, 2022
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Outdated suggestions cannot be applied.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.
PR types
Performance optimization
PR changes
Models
Description
优化transformer动态图fp16速度
1、优化点：
Adam优化器使用multi_tensor策略
clear_grad使用set_to_zero=False策略
dataloader num_workers支持>0 ：--num_workers
优化一些代码调用顺序
2、性能测试：