.

modal-labs · Aug 2, 2023 · 9a9fa29 · 9a9fa29
1 parent c15f360
commit 9a9fa29
Showing 1 changed file with 3 additions and 4 deletions.
diff --git a/06_gpu_and_ml/flan_t5/flan_t5_finetune.py b/06_gpu_and_ml/flan_t5/flan_t5_finetune.py
@@ -131,8 +131,9 @@ def on_save(self, args, state, control, **kwargs):
             """
             Event called after a checkpoint save.
             """
-            print("running commit on modal.Volume after model checkpoint")
-            self.volume.commit()
+            if state.is_world_process_zero:
+                print("running commit on modal.Volume after model checkpoint")
+                self.volume.commit()
 
     training_args = Seq2SeqTrainingArguments(
         # Save checkpoints to the mounted volume
@@ -142,8 +143,6 @@ def on_save(self, args, state, control, **kwargs):
         predict_with_generate=True,
         learning_rate=3e-5,
         num_train_epochs=num_train_epochs,
-        # Save logs to the mounted volume
-        logging_dir=str(VOL_MOUNT_PATH / "logs"),
         logging_strategy="steps",
         logging_steps=100,
         evaluation_strategy="epoch",