remove check for main process for trackers initialization (#18706)

examples/pytorch/image-classification/run_image_classification_no_trainer.py

@@ -414,14 +414,12 @@ def main():
         checkpointing_steps = None
 
     # We need to initialize the trackers we use, and also store our configuration.
-    # We initialize the trackers only on main process because `accelerator.log`
-    # only logs on main process and we don't want empty logs/runs on other processes.
+    # The trackers initializes automatically on the main process.
     if args.with_tracking:
-        if accelerator.is_main_process:
-            experiment_config = vars(args)
-            # TensorBoard cannot log Enums, need the raw value
-            experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
-            accelerator.init_trackers("image_classification_no_trainer", experiment_config)
+        experiment_config = vars(args)
+        # TensorBoard cannot log Enums, need the raw value
+        experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
+        accelerator.init_trackers("image_classification_no_trainer", experiment_config)
 
     # Get the metric function
     metric = evaluate.load("accuracy")

examples/pytorch/language-modeling/run_clm_no_trainer.py

@@ -516,14 +516,12 @@ def main():
         checkpointing_steps = None
 
     # We need to initialize the trackers we use, and also store our configuration.
-    # We initialize the trackers only on main process because `accelerator.log`
-    # only logs on main process and we don't want empty logs/runs on other processes.
+    # The trackers initializes automatically on the main process.
     if args.with_tracking:
-        if accelerator.is_main_process:
-            experiment_config = vars(args)
-            # TensorBoard cannot log Enums, need the raw value
-            experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
-            accelerator.init_trackers("clm_no_trainer", experiment_config)
+        experiment_config = vars(args)
+        # TensorBoard cannot log Enums, need the raw value
+        experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
+        accelerator.init_trackers("clm_no_trainer", experiment_config)
 
     # Train!
     total_batch_size = args.per_device_train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps

examples/pytorch/language-modeling/run_mlm_no_trainer.py

@@ -560,14 +560,12 @@ def main():
         checkpointing_steps = None
 
     # We need to initialize the trackers we use, and also store our configuration.
-    # We initialize the trackers only on main process because `accelerator.log`
-    # only logs on main process and we don't want empty logs/runs on other processes.
+    # The trackers initializes automatically on the main process.
     if args.with_tracking:
-        if accelerator.is_main_process:
-            experiment_config = vars(args)
-            # TensorBoard cannot log Enums, need the raw value
-            experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
-            accelerator.init_trackers("mlm_no_trainer", experiment_config)
+        experiment_config = vars(args)
+        # TensorBoard cannot log Enums, need the raw value
+        experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
+        accelerator.init_trackers("mlm_no_trainer", experiment_config)
 
     # Train!
     total_batch_size = args.per_device_train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps

examples/pytorch/multiple-choice/run_swag_no_trainer.py

@@ -513,14 +513,12 @@ def main():
         checkpointing_steps = None
 
     # We need to initialize the trackers we use, and also store our configuration.
-    # We initialize the trackers only on main process because `accelerator.log`
-    # only logs on main process and we don't want empty logs/runs on other processes.
+    # The trackers initializes automatically on the main process.
     if args.with_tracking:
-        if accelerator.is_main_process:
-            experiment_config = vars(args)
-            # TensorBoard cannot log Enums, need the raw value
-            experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
-            accelerator.init_trackers("swag_no_trainer", experiment_config)
+        experiment_config = vars(args)
+        # TensorBoard cannot log Enums, need the raw value
+        experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
+        accelerator.init_trackers("swag_no_trainer", experiment_config)
 
     # Metrics
     metric = evaluate.load("accuracy")

examples/pytorch/question-answering/run_qa_no_trainer.py

@@ -787,14 +787,12 @@ def main():
         checkpointing_steps = None
 
     # We need to initialize the trackers we use, and also store our configuration.
-    # We initialize the trackers only on main process because `accelerator.log`
-    # only logs on main process and we don't want empty logs/runs on other processes.
+    # The trackers initializes automatically on the main process.
     if args.with_tracking:
-        if accelerator.is_main_process:
-            experiment_config = vars(args)
-            # TensorBoard cannot log Enums, need the raw value
-            experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
-            accelerator.init_trackers("qa_no_trainer", experiment_config)
+        experiment_config = vars(args)
+        # TensorBoard cannot log Enums, need the raw value
+        experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
+        accelerator.init_trackers("qa_no_trainer", experiment_config)
 
     # Train!
     total_batch_size = args.per_device_train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps

examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py

@@ -512,14 +512,12 @@ def main():
     metric = evaluate.load("mean_iou")
 
     # We need to initialize the trackers we use, and also store our configuration.
-    # We initialize the trackers only on main process because `accelerator.log`
-    # only logs on main process and we don't want empty logs/runs on other processes.
+    # The trackers initializes automatically on the main process.
     if args.with_tracking:
-        if accelerator.is_main_process:
-            experiment_config = vars(args)
-            # TensorBoard cannot log Enums, need the raw value
-            experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
-            accelerator.init_trackers("semantic_segmentation_no_trainer", experiment_config)
+        experiment_config = vars(args)
+        # TensorBoard cannot log Enums, need the raw value
+        experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
+        accelerator.init_trackers("semantic_segmentation_no_trainer", experiment_config)
 
     # Train!
     total_batch_size = args.per_device_train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps

examples/pytorch/summarization/run_summarization_no_trainer.py

@@ -581,14 +581,12 @@ def main():
         checkpointing_steps = None
 
     # We need to initialize the trackers we use, and also store our configuration.
-    # We initialize the trackers only on main process because `accelerator.log`
-    # only logs on main process and we don't want empty logs/runs on other processes.
+    # The trackers initializes automatically on the main process.
     if args.with_tracking:
-        if accelerator.is_main_process:
-            experiment_config = vars(args)
-            # TensorBoard cannot log Enums, need the raw value
-            experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
-            accelerator.init_trackers("summarization_no_trainer", experiment_config)
+        experiment_config = vars(args)
+        # TensorBoard cannot log Enums, need the raw value
+        experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
+        accelerator.init_trackers("summarization_no_trainer", experiment_config)
 
     # Metric
     metric = evaluate.load("rouge")

examples/pytorch/text-classification/run_glue_no_trainer.py

@@ -459,14 +459,12 @@ def main():
         checkpointing_steps = None
 
     # We need to initialize the trackers we use, and also store our configuration.
-    # We initialize the trackers only on main process because `accelerator.log`
-    # only logs on main process and we don't want empty logs/runs on other processes.
+    # The trackers initializes automatically on the main process.
     if args.with_tracking:
-        if accelerator.is_main_process:
-            experiment_config = vars(args)
-            # TensorBoard cannot log Enums, need the raw value
-            experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
-            accelerator.init_trackers("glue_no_trainer", experiment_config)
+        experiment_config = vars(args)
+        # TensorBoard cannot log Enums, need the raw value
+        experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
+        accelerator.init_trackers("glue_no_trainer", experiment_config)
 
     # Get the metric function
     if args.task_name is not None:

examples/pytorch/token-classification/run_ner_no_trainer.py

@@ -574,14 +574,12 @@ def main():
         checkpointing_steps = None
 
     # We need to initialize the trackers we use, and also store our configuration.
-    # We initialize the trackers only on main process because `accelerator.log`
-    # only logs on main process and we don't want empty logs/runs on other processes.
+    # The trackers initializes automatically on the main process.
     if args.with_tracking:
-        if accelerator.is_main_process:
-            experiment_config = vars(args)
-            # TensorBoard cannot log Enums, need the raw value
-            experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
-            accelerator.init_trackers("ner_no_trainer", experiment_config)
+        experiment_config = vars(args)
+        # TensorBoard cannot log Enums, need the raw value
+        experiment_config["lr_scheduler_type"] = experiment_config["lr_scheduler_type"].value
+        accelerator.init_trackers("ner_no_trainer", experiment_config)
 
     # Metrics
     metric = evaluate.load("seqeval")