Akoumparouli/nemo ux update param name (NVIDIA#10441)

* NeMoLogger: update dir to log_dir

Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>

* NeMologger: update calls

Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>

---------

Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>
Co-authored-by: Marc Romeyn <mromeijn@nvidia.com>
Signed-off-by: Lifu Zhang <tomzhanglf@gmail.com>

examples/llm/megatron_gpt_pretraining.py

@@ -102,7 +102,7 @@ def get_args():
     )
 
     nemo_logger = NeMoLogger(
-        dir=args.experiment_dir,
+        log_dir=args.experiment_dir,
     )
 
     train(

nemo/collections/llm/recipes/log/default.py

@@ -60,7 +60,7 @@ def default_log(
         name=name,
         tensorboard=tensorboard_logger,
         wandb=wandb_logger,
-        dir=dir,
+        log_dir=dir,
     )
 
 

nemo/collections/llm/tools/auto_configurator/core/base_config.py

@@ -203,7 +203,7 @@ def get_logger(self) -> Config[nl.NeMoLogger]:
             ckpt=ckpt,
             tensorboard=tb_logger,
             wandb=None,
-            dir=self.config.path_to_logs,
+            log_dir=self.config.path_to_logs,
         )
 
     def get_run_config(self) -> dict:

nemo/lightning/nemo_logger.py

@@ -35,7 +35,7 @@ class NeMoLogger(IOMixin):
 
     Args:
         name (str): Name of the experiment.
-        dir (Optional[str]): Directory to save logs.
+        log_dir (Optional[str]): Directory to save logs.
         explicit_log_dir (Optional[str]): Explicit log directory.
         version (Optional[str]): Version of the experiment.
         use_datetime_version (bool): Whether to use datetime as version.
@@ -56,7 +56,7 @@ class NeMoLogger(IOMixin):
     """
 
     name: str = "default"
-    dir: Optional[str] = None
+    log_dir: Optional[str] = None
     explicit_log_dir: Optional[str] = None
     version: Optional[str] = None
     use_datetime_version: bool = True
@@ -99,9 +99,9 @@ def setup(self, trainer: Union[pl.Trainer, fl.Fabric], resume_if_exists: bool =
                     f"that was passed to nemo_logger container a logger, but update_logger_directory is False. This means "
                     f"that the trainer's logger directory may not match with the explicit_log_dir."
                 )
-            if self.dir or self.version:
+            if self.log_dir or self.version:
                 logging.error(
-                    f"nemo logger received explicit_log_dir: {self.explicit_log_dir} and at least one of dir: {self.dir}, "
+                    f"nemo logger received explicit_log_dir: {self.explicit_log_dir} and at least one of dir: {self.log_dir}, "
                     f"or version: {self.version}. Please note that dir, name, and version will be ignored."
                 )
             if is_global_rank_zero() and Path(self.explicit_log_dir).exists():
@@ -110,8 +110,8 @@ def setup(self, trainer: Union[pl.Trainer, fl.Fabric], resume_if_exists: bool =
 
         else:
             # Default dir to ./nemo_experiments if None was passed
-            _dir = self.dir
-            if self.dir is None:
+            _dir = self.log_dir
+            if self.log_dir is None:
                 _dir = str(Path.cwd() / "nemo_experiments")
 
             if not self.name:

tests/collections/llm/auto_conf/test_base_configs.py

@@ -118,7 +118,7 @@ def get_logger() -> run.Config[nl.NeMoLogger]:
         ckpt=ckpt,
         tensorboard=tb_logger,
         wandb=None,
-        dir="/",
+        log_dir="/",
     )
 
 

tests/collections/llm/gpt/model/megatron_ssm_pretraining.py

@@ -116,7 +116,7 @@ def get_args():
     )
 
     nemo_logger = NeMoLogger(
-        dir=args.experiment_dir,
+        log_dir=args.experiment_dir,
     )
 
     train(

tests/collections/llm/test_mnist_model_nemo2.py

@@ -509,7 +509,7 @@ def run_train_mnist_litautoencoder_with_megatron_strategy_single_gpu():
             tb_logger = TensorBoardLogger(save_dir=str(save_dir), name=name)
             # Setup the logger and train the model
             nemo_logger = NeMoLogger(
-                dir=str(root_dir),  # WARNING: passing a path in here results in mutating the Path class.
+                log_dir=str(root_dir),  # WARNING: passing a path in here results in mutating the Path class.
                 name=name,
                 tensorboard=tb_logger,
                 ckpt=checkpoint_callback,

tests/collections/llm/test_mnist_model_nemo2_fsdp.py

@@ -531,7 +531,7 @@ def run_train_mnist_litautoencoder_with_fsdp_strategy_single_gpu():
             tb_logger = TensorBoardLogger(save_dir=str(save_dir), name=name)
             # Setup the logger and train the model
             nemo_logger = NeMoLogger(
-                dir=str(root_dir),  # WARNING: passing a path in here results in mutating the Path class.
+                log_dir=str(root_dir),  # WARNING: passing a path in here results in mutating the Path class.
                 name=name,
                 tensorboard=tb_logger,
                 ckpt=checkpoint_callback,

tests/lightning/test_nemo_logger.py

@@ -180,7 +180,7 @@ def test_resume(self, trainer, tmp_path):
 
         logger = nl.NeMoLogger(
             name="default",
-            dir=str(tmp_path) + "/test_resume",
+            log_dir=str(tmp_path) + "/test_resume",
             version="version_0",
             use_datetime_version=False,
         )