Will this work?

Author: johnml1135

machine/jobs/settings.yaml

@@ -7,7 +7,7 @@ default:
     parent_model_name: facebook/nllb-200-distilled-1.3B
     train_params:
       do_train: true
-      optim: adamw_torch
+      optim: apollo_adamw
       warmup_steps: 1000
       per_device_train_batch_size: 16
       gradient_accumulation_steps: 4