Address feedback

Signed-off-by: SeanNaren <[email protected]>
NVIDIA · Feb 9, 2023 · 841f665 · 841f665
1 parent 0cce224
commit 841f665
Show file tree

Hide file tree

Showing 4 changed files with 5 additions and 7 deletions.
diff --git a/nemo/utils/cloud.py b/nemo/utils/cloud.py
@@ -136,7 +136,6 @@ def __new_hash__(self):
 
         if dist.get_local_rank():
             _install_system_libraries()
-        print("WORLD SIZE NEMO", dist.get_world_size())
         return dist.barrier()  # wait for main process
     _install_system_libraries()
     return
diff --git a/tutorials/cloud/README.md b/tutorials/cloud/README.md
@@ -2,7 +2,6 @@
 
 The best way to get started with NeMo in the cloud.
 
-| Domain | Tutorial                                          | Key Words              | Github URL                                                                                  |
-|--------|---------------------------------------------------|------------------------|---------------------------------------------------------------------------------------------|
-| ASR | Quickstart Training an ASR model on AWS SageMaker | ASR, Basics            | [NeMo ASR - Quickstart Cloud Training](SageMaker_ASR_Training.ipynb)                        |
-| ASR | Multi-node ASR fine-tuning on AWS SageMaker       | ASR, DDP, Common Voice | [NeMo ASR - Cloud Multi-node Fine-tuning](ASR_Finetuning_at_Scale_with_AWS_SageMaker.ipynb) |
+## ASR
+* [Quickstart: Training an ASR model on AWS SageMaker](aws/SageMaker_ASR_Training.ipynb)
+* [Multi-Node Scaling: ASR Fine-Tuning on AWS SageMaker](aws/ASR_Finetuning_at_Scale_with_AWS_SageMaker.ipynb)
diff --git a/tutorials/cloud/aws/ASR_Finetuning_at_Scale_with_AWS_SageMaker.ipynb b/tutorials/cloud/aws/ASR_Finetuning_at_Scale_with_AWS_SageMaker.ipynb
@@ -570,7 +570,7 @@
     "conf.trainer.strategy = \"smddp\"\n",
     "conf.trainer.max_epochs = 300\n",
     "conf.trainer.accumulate_grad_batches = accumulate_grad_batches\n",
-    "conf.trainer.precision = 16\n",
+    "conf.trainer.precision = 16 # if using Ampere or above (i.e ml.p4d.24xlarge), you can use bf16 (BFloat16) \n",
     "\n",
     "# resume flags if crashes occur\n",
     "conf.exp_manager.resume_if_exists=True\n",

diff --git a/tutorials/cloud/aws/SageMaker_ASR_Training.ipynb b/tutorials/cloud/aws/SageMaker_ASR_Training.ipynb
@@ -297,7 +297,7 @@
     "conf.model.spec_augment.time_masks = 5\n",
     "\n",
     "# Set Optimizer parameters\n",
-    "conf.model.optim.lr = 2.0\n",
+    "conf.model.optim.lr = 2.0 # by default we using Noam scheduling, the LR is a multiplier \n",
     "\n",
     "OmegaConf.save(conf, config_dir / 'config.yaml')"
    ]