Skip to content
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

update for tf2.4 #908

Merged
merged 8 commits into from
Dec 27, 2020
Merged

update for tf2.4 #908

Changes from 1 commit
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
da3171b
update for tf2.4
fsx950223 Dec 24, 2020
1cfb9f2
fix mixed precision with recompute gradient
fsx950223 Dec 24, 2020
750b36d
update README
fsx950223 Dec 24, 2020
aa70c66
fix multi gpus training
fsx950223 Dec 25, 2020
4dd12f6
update README
fsx950223 Dec 25, 2020
e771e73
fix LossScaleOptimizer bug
fsx950223 Dec 25, 2020
7f2423e
disable steps_per_execution in default
fsx950223 Dec 25, 2020
77b627a
split all reduce
fsx950223 Dec 25, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 0 additions & 2 deletions efficientdet/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -622,13 +622,11 @@ def build_model_with_precision(pp, mm, ii, *args, **kwargs):
inputs = tf.cast(ii, tf.bfloat16)
with tf.tpu.bfloat16_scope():
outputs = mm(inputs, *args, **kwargs)
set_precision_policy('float32')
Copy link
Collaborator Author

@fsx950223 fsx950223 Dec 24, 2020 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

After remove 2 lines, I could train estimator model with recompute_grad and mixed_precision.
Why set policy back to float32? Could I remove them? @mingxingtan

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, please feel free to remove it. it is not necessary.

elif pp == 'mixed_float16':
set_precision_policy(pp)
inputs = tf.cast(ii, tf.float16)
with float16_scope():
outputs = mm(inputs, *args, **kwargs)
set_precision_policy('float32')
elif not pp or pp == 'float32':
outputs = mm(ii, *args, **kwargs)
else:
Expand Down