Skip to content

mixed precision fix for runner #292

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

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

mixed precision fix for runner #292

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
9 changes: 8 additions & 1 deletion tensorflow_gnn/runner/orchestration.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -315,13 +315,20 @@ def apply_fn(ds,
drop_remainder,
global_batch_size)

def adapted_model_fn():
def adapted_model_fn(options = None):
if isinstance(preprocess_model.output, collections.abc.Sequence):
x, *_ = preprocess_model.output
else:
x = preprocess_model.output
m = task.adapt(model_fn(x.spec))
optimizer = optimizer_fn()
if options and options.policy:
# Cast logits to `tf.keras.backend.floatx()` for mixed_precision.
# For more details, see:
# https://www.tensorflow.org/guide/mixed_precision#building_the_model.
floatx = tf.keras.backend.floatx()
outputs = [tf.cast(o, dtype=floatx) for o in m.outputs]
m = tf.keras.Model(m.inputs, outputs)
if train_padding is None:
m.compile(optimizer, loss=task.losses(), metrics=task.metrics())
else:
Expand Down
8 changes: 0 additions & 8 deletions tensorflow_gnn/runner/trainers/keras_fit.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -269,14 +269,6 @@ def per_replica_ds_fn(input_context, *, delegate, repeat):
with self._strategy.scope():
model = model_fn()

if self._options and self._options.policy:
# Cast logits to `tf.keras.backend.floatx()` for mixed_precision.
# For more details, see:
# https://www.tensorflow.org/guide/mixed_precision#building_the_model.
floatx = tf.keras.backend.floatx()
outputs = [tf.cast(o, dtype=floatx) for o in model.outputs]
model = tf.keras.Model(model.inputs, outputs)

model.fit(
train_ds,
epochs=epochs,
Expand Down