NeMo Megatron launcher and tools. Contribute to roclark/NeMo-Megatron-Launcher development by creating an account on GitHub.
2 changes: 1 addition & 1 deletion 2 launcher_scripts/conf/training/mt5/11b.yaml @@ -54,7 +54,7 @@ exp_manager: model: # model parallelism micro_batch_size: 8 micro_batch_size: 24 global_batch_size: 1920 # will use more micro batches to reach global batch size tensor_model_...
NeMo Megatron launcher and tools. Contribute to roclark/NeMo-Megatron-Launcher development by creating an account on GitHub.