Skip to content

GitLab

"git@developer.sourcefind.cn:modelzoo/solov2-pytorch.git" did not exist on "7c2b8148c8b2e8b4fa4d0692055153e5f4449072"
Commit 00c24fde authored by Jinoo Baek's avatar Jinoo Baek Committed by A. Unique TensorFlower
Browse files

Indentation bug. Divide by num_replicas_in_sync once. 

PiperOrigin-RevId: 448508763
parent 069ad593
Hide whitespace changes
Inline Side-by-side
Showing with 4 additions and 4 deletions
official/modeling/multitask/multitask.py
View file @ 00c24fde
...@@ -138,10 +138,10 @@ class MultiTask(tf.Module, metaclass=abc.ABCMeta): ...@@ -138,10 +138,10 @@ class MultiTask(tf.Module, metaclass=abc.ABCMeta):
self.tasks[name].process_metrics(task_metrics[name], labels, outputs, self.tasks[name].process_metrics(task_metrics[name], labels, outputs,
**kwargs) **kwargs)
# Scales loss as the default gradients allreduce performs sum inside # Scales loss as the default gradients allreduce performs sum inside
# the optimizer. # the optimizer.
scaled_loss = total_loss / tf.distribute.get_strategy( scaled_loss = total_loss / tf.distribute.get_strategy(
).num_replicas_in_sync ).num_replicas_in_sync
tvars = multi_task_model.trainable_variables tvars = multi_task_model.trainable_variables
grads = tape.gradient(scaled_loss, tvars) grads = tape.gradient(scaled_loss, tvars)
optimizer.apply_gradients(list(zip(grads, tvars))) optimizer.apply_gradients(list(zip(grads, tvars)))
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment