Add examples telemetry (#17552)

* Add examples telemetry

* Alternative approach

* Add to all other examples

* Add to templates as well

* Put framework separately

* Same for TensorFlow

examples/flax/image-captioning/run_image_captioning_flax.py

@@ -52,7 +52,7 @@ from transformers import (
     HfArgumentParser,
     is_tensorboard_available,
 )
-from transformers.utils import get_full_repo_name, is_offline_mode
+from transformers.utils import get_full_repo_name, is_offline_mode, send_example_telemetry
 
 
 logger = logging.getLogger(__name__)
@@ -388,6 +388,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_image_captioning", model_args, data_args, framework="flax")
+
     if (
         os.path.exists(training_args.output_dir)
         and os.listdir(training_args.output_dir)

examples/flax/language-modeling/run_clm_flax.py

@@ -58,7 +58,7 @@ from transformers import (
     set_seed,
 )
 from transformers.testing_utils import CaptureLogger
-from transformers.utils import get_full_repo_name
+from transformers.utils import get_full_repo_name, send_example_telemetry
 
 
 logger = logging.getLogger(__name__)
@@ -328,6 +328,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_clm", model_args, data_args, framework="flax")
+
     if (
         os.path.exists(training_args.output_dir)
         and os.listdir(training_args.output_dir)

examples/flax/language-modeling/run_mlm_flax.py

@@ -58,7 +58,7 @@ from transformers import (
     is_tensorboard_available,
     set_seed,
 )
-from transformers.utils import get_full_repo_name
+from transformers.utils import get_full_repo_name, send_example_telemetry
 
 
 MODEL_CONFIG_CLASSES = list(FLAX_MODEL_FOR_MASKED_LM_MAPPING.keys())
@@ -365,6 +365,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_mlm", model_args, data_args, framework="flax")
+
     if (
         os.path.exists(training_args.output_dir)
         and os.listdir(training_args.output_dir)

examples/flax/language-modeling/run_t5_mlm_flax.py

@@ -57,7 +57,7 @@ from transformers import (
     set_seed,
 )
 from transformers.models.t5.modeling_flax_t5 import shift_tokens_right
-from transformers.utils import get_full_repo_name
+from transformers.utils import get_full_repo_name, send_example_telemetry
 
 
 MODEL_CONFIG_CLASSES = list(FLAX_MODEL_FOR_MASKED_LM_MAPPING.keys())
@@ -498,6 +498,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_t5_mlm", model_args, data_args, framework="flax")
+
     if (
         os.path.exists(training_args.output_dir)
         and os.listdir(training_args.output_dir)

examples/flax/question-answering/run_qa.py

@@ -53,7 +53,7 @@ from transformers import (
     PreTrainedTokenizerFast,
     is_tensorboard_available,
 )
-from transformers.utils import check_min_version, get_full_repo_name
+from transformers.utils import check_min_version, get_full_repo_name, send_example_telemetry
 from utils_qa import postprocess_qa_predictions
 
 
@@ -424,6 +424,10 @@ def main():
         model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_qa", model_args, data_args, framework="flax")
     # endregion
 
     # region Logging

examples/flax/summarization/run_summarization_flax.py

@@ -54,7 +54,7 @@ from transformers import (
     HfArgumentParser,
     is_tensorboard_available,
 )
-from transformers.utils import get_full_repo_name, is_offline_mode
+from transformers.utils import get_full_repo_name, is_offline_mode, send_example_telemetry
 
 
 logger = logging.getLogger(__name__)
@@ -399,6 +399,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_summarization", model_args, data_args, framework="flax")
+
     if (
         os.path.exists(training_args.output_dir)
         and os.listdir(training_args.output_dir)

examples/flax/text-classification/run_flax_glue.py

@@ -48,7 +48,7 @@ from transformers import (
     TrainingArguments,
     is_tensorboard_available,
 )
-from transformers.utils import check_min_version, get_full_repo_name
+from transformers.utils import check_min_version, get_full_repo_name, send_example_telemetry
 
 
 logger = logging.getLogger(__name__)
@@ -308,6 +308,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_glue", model_args, data_args, framework="flax")
+
     # Make one log on every process with the configuration for debugging.
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/flax/token-classification/run_flax_ner.py

@@ -47,7 +47,7 @@ from transformers import (
     HfArgumentParser,
     is_tensorboard_available,
 )
-from transformers.utils import check_min_version, get_full_repo_name
+from transformers.utils import check_min_version, get_full_repo_name, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -366,6 +366,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_ner", model_args, data_args, framework="flax")
+
     # Make one log on every process with the configuration for debugging.
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/flax/vision/run_image_classification.py

@@ -53,7 +53,7 @@ from transformers import (
     is_tensorboard_available,
     set_seed,
 )
-from transformers.utils import get_full_repo_name
+from transformers.utils import get_full_repo_name, send_example_telemetry
 
 
 logger = logging.getLogger(__name__)
@@ -256,6 +256,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_image_classification", model_args, data_args, framework="flax")
+
     if (
         os.path.exists(training_args.output_dir)
         and os.listdir(training_args.output_dir)

examples/pytorch/audio-classification/run_audio_classification.py

@@ -37,7 +37,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -197,6 +197,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_audio_classification", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/contrastive-image-text/run_clip.py

@@ -47,7 +47,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -233,6 +233,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_clip", model_args, data_args)
+
     # 2. Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/image-classification/run_image_classification.py

@@ -45,7 +45,7 @@ from transformers import (
     TrainingArguments,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -175,6 +175,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_image_classification", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/image-classification/run_image_classification_no_trainer.py

@@ -47,7 +47,7 @@ from transformers import (
     SchedulerType,
     get_scheduler,
 )
-from transformers.utils import get_full_repo_name
+from transformers.utils import get_full_repo_name, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -201,6 +201,10 @@ def parse_args():
 def main():
     args = parse_args()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_image_classification_no_trainer", args)
+
     # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
     # If we're using tracking, we also need to initialize it here and it will by default pick up all supported trackers
     # in the environment

examples/pytorch/image-pretraining/run_mae.py

@@ -34,7 +34,7 @@ from transformers import (
     ViTMAEForPreTraining,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -175,6 +175,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_mae", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/image-pretraining/run_mim.py

@@ -37,7 +37,7 @@ from transformers import (
     TrainingArguments,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -239,6 +239,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_mim", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/language-modeling/run_clm.py

@@ -48,7 +48,7 @@ from transformers import (
 )
 from transformers.testing_utils import CaptureLogger
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -214,6 +214,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_clm", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/language-modeling/run_clm_no_trainer.py

@@ -52,7 +52,7 @@ from transformers import (
     default_data_collator,
     get_scheduler,
 )
-from transformers.utils import get_full_repo_name
+from transformers.utils import get_full_repo_name, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -239,6 +239,10 @@ def parse_args():
 def main():
     args = parse_args()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_clm_no_trainer", args)
+
     # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
     # If we're using tracking, we also need to initialize it here and it will by default pick up all supported trackers
     # in the environment

examples/pytorch/language-modeling/run_mlm.py

@@ -47,7 +47,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -224,6 +224,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_mlm", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/language-modeling/run_mlm_no_trainer.py

@@ -52,7 +52,7 @@ from transformers import (
     SchedulerType,
     get_scheduler,
 )
-from transformers.utils import get_full_repo_name
+from transformers.utils import get_full_repo_name, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -248,6 +248,10 @@ def parse_args():
 def main():
     args = parse_args()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_mlm_no_trainer", args)
+
     # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
     # If we're using tracking, we also need to initialize it here and it will by default pick up all supported trackers
     # in the environment

examples/pytorch/language-modeling/run_plm.py

@@ -42,7 +42,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -220,6 +220,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_plm", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/multiple-choice/run_swag.py

@@ -43,7 +43,7 @@ from transformers import (
 )
 from transformers.tokenization_utils_base import PreTrainedTokenizerBase
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import PaddingStrategy, check_min_version
+from transformers.utils import PaddingStrategy, check_min_version, send_example_telemetry
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
@@ -225,6 +225,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_swag", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/multiple-choice/run_swag_no_trainer.py

@@ -51,7 +51,7 @@ from transformers import (
     default_data_collator,
     get_scheduler,
 )
-from transformers.utils import PaddingStrategy, get_full_repo_name
+from transformers.utils import PaddingStrategy, get_full_repo_name, send_example_telemetry
 
 
 logger = get_logger(__name__)
@@ -273,6 +273,10 @@ class DataCollatorForMultipleChoice:
 def main():
     args = parse_args()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_swag_no_trainer", args)
+
     # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
     # If we're using tracking, we also need to initialize it here and it will by default pick up all supported trackers
     # in the environment

examples/pytorch/question-answering/run_qa.py

@@ -42,7 +42,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 from utils_qa import postprocess_qa_predictions
 
@@ -226,6 +226,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_qa", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/question-answering/run_qa_beam_search.py

@@ -41,7 +41,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 from utils_qa import postprocess_qa_predictions_with_beam_search
 
@@ -225,6 +225,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_qa_beam_search", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/question-answering/run_qa_beam_search_no_trainer.py

@@ -49,7 +49,7 @@ from transformers import (
     default_data_collator,
     get_scheduler,
 )
-from transformers.utils import check_min_version, get_full_repo_name
+from transformers.utils import check_min_version, get_full_repo_name, send_example_telemetry
 from transformers.utils.versions import require_version
 from utils_qa import postprocess_qa_predictions_with_beam_search
 
@@ -291,6 +291,10 @@ def parse_args():
 def main():
     args = parse_args()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_qa_beam_search_no_trainer", args)
+
     # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
     # If we're using tracking, we also need to initialize it here and it will pick up all supported trackers in the environment
     accelerator = Accelerator(log_with="all", logging_dir=args.output_dir) if args.with_tracking else Accelerator()

examples/pytorch/question-answering/run_qa_no_trainer.py

@@ -50,7 +50,7 @@ from transformers import (
     default_data_collator,
     get_scheduler,
 )
-from transformers.utils import check_min_version, get_full_repo_name
+from transformers.utils import check_min_version, get_full_repo_name, send_example_telemetry
 from transformers.utils.versions import require_version
 from utils_qa import postprocess_qa_predictions
 
@@ -329,6 +329,10 @@ def parse_args():
 def main():
     args = parse_args()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_qa_no_trainer", args)
+
     # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
     # If we're using tracking, we also need to initialize it here and it will by default pick up all supported trackers
     # in the environment

examples/pytorch/question-answering/run_seq2seq_qa.py

@@ -39,7 +39,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import EvalLoopOutput, EvalPrediction, get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -271,6 +271,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_seq2seq_qa", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/semantic-segmentation/run_semantic_segmentation.py

@@ -42,7 +42,7 @@ from transformers import (
     default_data_collator,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -266,6 +266,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_semantic_segmentation", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/semantic-segmentation/run_semantic_segmentation_no_trainer.py

@@ -44,7 +44,7 @@ from transformers import (
     default_data_collator,
     get_scheduler,
 )
-from transformers.utils import get_full_repo_name
+from transformers.utils import get_full_repo_name, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -315,6 +315,10 @@ def parse_args():
 def main():
     args = parse_args()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_semantic_segmentation_no_trainer", args)
+
     # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
     # If we're using tracking, we also need to initialize it here and it will by default pick up all supported trackers
     # in the environment

examples/pytorch/speech-pretraining/run_wav2vec2_pretraining_no_trainer.py

@@ -43,7 +43,7 @@ from transformers import (
     set_seed,
 )
 from transformers.models.wav2vec2.modeling_wav2vec2 import _compute_mask_indices, _sample_negative_indices
-from transformers.utils import get_full_repo_name
+from transformers.utils import get_full_repo_name, send_example_telemetry
 
 
 logger = get_logger(__name__)
@@ -363,6 +363,10 @@ def main():
     # We now keep distinct sets of args, for a cleaner separation of concerns.
     args = parse_args()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_wav2vec2_pretraining_no_trainer", args)
+
     # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
     accelerator = Accelerator()
     logger.info(accelerator.state, main_process_only=False)

examples/pytorch/speech-recognition/run_speech_recognition_ctc.py

@@ -44,7 +44,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint, is_main_process
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -376,6 +376,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_speech_recognition_ctc", model_args, data_args)
+
     # Detecting last checkpoint.
     last_checkpoint = None
     if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:

examples/pytorch/speech-recognition/run_speech_recognition_seq2seq.py

@@ -42,7 +42,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint, is_main_process
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -239,6 +239,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_speech_recognition_seq2seq", model_args, data_args)
+
     # 2. Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/summarization/run_summarization.py

@@ -46,7 +46,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version, is_offline_mode
+from transformers.utils import check_min_version, is_offline_mode, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -302,6 +302,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_summarization", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/summarization/run_summarization_no_trainer.py

@@ -50,7 +50,7 @@ from transformers import (
     SchedulerType,
     get_scheduler,
 )
-from transformers.utils import get_full_repo_name, is_offline_mode
+from transformers.utils import get_full_repo_name, is_offline_mode, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -319,6 +319,10 @@ def parse_args():
 
 def main():
     args = parse_args()
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_summarization_no_trainer", args)
+
     # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
     # If we're using tracking, we also need to initialize it here and it will by default pick up all supported trackers
     # in the environment

examples/pytorch/text-classification/run_glue.py

@@ -42,7 +42,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -215,6 +215,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_glue", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/text-classification/run_glue_no_trainer.py

@@ -42,7 +42,7 @@ from transformers import (
     default_data_collator,
     get_scheduler,
 )
-from transformers.utils import get_full_repo_name
+from transformers.utils import get_full_repo_name, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -205,6 +205,9 @@ def parse_args():
 
 def main():
     args = parse_args()
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_glue_no_trainer", args)
 
     # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
     # If we're using tracking, we also need to initialize it here and it will by default pick up all supported trackers

examples/pytorch/text-classification/run_xnli.py

@@ -42,7 +42,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -112,8 +112,6 @@ class DataTrainingArguments:
             )
         },
     )
-    server_ip: Optional[str] = field(default=None, metadata={"help": "For distant debugging."})
-    server_port: Optional[str] = field(default=None, metadata={"help": "For distant debugging."})
 
 
 @dataclass
@@ -176,14 +174,9 @@ def main():
     parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
     model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
-    # Setup distant debugging if needed
-    if data_args.server_ip and data_args.server_port:
-        # Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script
-        import ptvsd
-
-        print("Waiting for debugger attach")
-        ptvsd.enable_attach(address=(data_args.server_ip, data_args.server_port), redirect_output=True)
-        ptvsd.wait_for_attach()
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_xnli", model_args)
 
     # Setup logging
     logging.basicConfig(

examples/pytorch/token-classification/run_ner.py

@@ -43,7 +43,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -216,6 +216,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_ner", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/token-classification/run_ner_no_trainer.py

@@ -49,7 +49,7 @@ from transformers import (
     default_data_collator,
     get_scheduler,
 )
-from transformers.utils import get_full_repo_name
+from transformers.utils import get_full_repo_name, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -259,6 +259,10 @@ def parse_args():
 def main():
     args = parse_args()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_ner_no_trainer", args)
+
     # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
     # If we're using tracking, we also need to initialize it here and it will by default pick up all supported trackers
     # in the environment

examples/pytorch/translation/run_translation.py

@@ -46,7 +46,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -260,6 +260,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_translation", model_args, data_args)
+
     # Setup logging
     logging.basicConfig(
         format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",

examples/pytorch/translation/run_translation_no_trainer.py

@@ -51,7 +51,7 @@ from transformers import (
     default_data_collator,
     get_scheduler,
 )
-from transformers.utils import get_full_repo_name
+from transformers.utils import get_full_repo_name, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -305,6 +305,10 @@ def main():
     # Parse the arguments
     args = parse_args()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_translation_no_trainer", args)
+
     # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
     # If we're using tracking, we also need to initialize it here and it will by default pick up all supported trackers
     # in the environment

examples/tensorflow/language-modeling/run_clm.py

@@ -53,6 +53,7 @@ from transformers import (
     create_optimizer,
     set_seed,
 )
+from transformers.utils import send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -232,6 +233,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_clm", model_args, data_args, framework="tensorflow")
+
     # Sanity checks
     if data_args.dataset_name is None and data_args.train_file is None and data_args.validation_file is None:
         raise ValueError("Need either a dataset name or a training/validation file.")

examples/tensorflow/language-modeling/run_mlm.py

@@ -55,6 +55,7 @@ from transformers import (
     create_optimizer,
     set_seed,
 )
+from transformers.utils import send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -242,6 +243,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_mlm", model_args, data_args, framework="tensorflow")
+
     # Sanity checks
     if data_args.dataset_name is None and data_args.train_file is None and data_args.validation_file is None:
         raise ValueError("Need either a dataset name or a training/validation file.")

examples/tensorflow/multiple-choice/run_swag.py

@@ -44,7 +44,7 @@ from transformers import (
     set_seed,
 )
 from transformers.tokenization_utils_base import PreTrainedTokenizerBase
-from transformers.utils import PaddingStrategy, check_min_version
+from transformers.utils import PaddingStrategy, check_min_version, send_example_telemetry
 
 
 # Will error if the minimal version of Transformers is not installed. Remove at your own risks.
@@ -246,6 +246,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_swag", model_args, data_args, framework="tensorflow")
+
     output_dir = Path(training_args.output_dir)
     output_dir.mkdir(parents=True, exist_ok=True)
     # endregion

examples/tensorflow/question-answering/run_qa.py

@@ -41,7 +41,7 @@ from transformers import (
     TFTrainingArguments,
     set_seed,
 )
-from transformers.utils import CONFIG_NAME, TF2_WEIGHTS_NAME, check_min_version
+from transformers.utils import CONFIG_NAME, TF2_WEIGHTS_NAME, check_min_version, send_example_telemetry
 from utils_qa import postprocess_qa_predictions
 
 
@@ -242,6 +242,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_qa", model_args, data_args, framework="tensorflow")
+
     output_dir = Path(training_args.output_dir)
     output_dir.mkdir(parents=True, exist_ok=True)
     # endregion

examples/tensorflow/summarization/run_summarization.py

@@ -44,7 +44,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version, is_offline_mode
+from transformers.utils import check_min_version, is_offline_mode, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -348,6 +348,10 @@ def main():
         model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_summarization", model_args, data_args, framework="tensorflow")
     # endregion
 
     # region Logging

examples/tensorflow/text-classification/run_glue.py

@@ -39,7 +39,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint, is_main_process
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 
 
 # region Helper functions
@@ -206,6 +206,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_glue", model_args, data_args, framework="tensorflow")
+
     if not (training_args.do_train or training_args.do_eval or training_args.do_predict):
         exit("Must specify at least one of --do_train, --do_eval or --do_predict!")
     # endregion

examples/tensorflow/text-classification/run_text_classification.py

@@ -37,7 +37,7 @@ from transformers import (
     TFTrainingArguments,
     set_seed,
 )
-from transformers.utils import CONFIG_NAME, TF2_WEIGHTS_NAME
+from transformers.utils import CONFIG_NAME, TF2_WEIGHTS_NAME, send_example_telemetry
 
 
 os.environ["TF_CPP_MIN_LOG_LEVEL"] = "1"  # Reduce the amount of console output from TF
@@ -196,6 +196,11 @@ def main():
         model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_text_classification", model_args, data_args, framework="tensorflow")
+
     output_dir = Path(training_args.output_dir)
     output_dir.mkdir(parents=True, exist_ok=True)
     # endregion

examples/tensorflow/token-classification/run_ner.py

@@ -41,6 +41,7 @@ from transformers import (
     create_optimizer,
     set_seed,
 )
+from transformers.utils import send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -252,6 +253,10 @@ def main():
     # region Argument Parsing
     parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TFTrainingArguments))
     model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_ner", model_args, data_args, framework="tensorflow")
     # endregion
 
     # region Setup logging

examples/tensorflow/translation/run_translation.py

@@ -47,7 +47,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
-from transformers.utils import check_min_version
+from transformers.utils import check_min_version, send_example_telemetry
 from transformers.utils.versions import require_version
 
 
@@ -318,6 +318,10 @@ def main():
         model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_translation", model_args, data_args, framework="tensorflow")
     # endregion
 
     # region Logging

src/transformers/utils/__init__.py

@@ -74,6 +74,7 @@ from .hub import (
     is_local_clone,
     is_offline_mode,
     is_remote_url,
+    send_example_telemetry,
     url_to_filename,
 )
 from .import_utils import (

src/transformers/utils/hub.py

@@ -109,6 +109,7 @@ if os.environ.get("HUGGINGFACE_CO_RESOLVE_ENDPOINT", None) is not None:
     HUGGINGFACE_CO_RESOLVE_ENDPOINT = os.environ.get("HUGGINGFACE_CO_RESOLVE_ENDPOINT", None)
 HUGGINGFACE_CO_RESOLVE_ENDPOINT = os.environ.get("HF_ENDPOINT", HUGGINGFACE_CO_RESOLVE_ENDPOINT)
 HUGGINGFACE_CO_PREFIX = HUGGINGFACE_CO_RESOLVE_ENDPOINT + "/{model_id}/resolve/{revision}/{filename}"
+HUGGINGFACE_CO_EXAMPLES_TELEMETRY = HUGGINGFACE_CO_RESOLVE_ENDPOINT + "/telemetry/examples"
 
 
 def is_remote_url(url_or_filename):
@@ -1028,3 +1029,41 @@ def get_full_repo_name(model_id: str, organization: Optional[str] = None, token:
         return f"{username}/{model_id}"
     else:
         return f"{organization}/{model_id}"
+
+
+def send_example_telemetry(example_name, *example_args, framework="pytorch"):
+    """
+    Sends telemetry that helps tracking the examples use.
+
+    Args:
+        example_name (`str`): The name of the example.
+        *example_args (dataclasses or `argparse.ArgumentParser`): The arguments to the script. This function will only
+            try to extract the model and dataset name from those. Nothing else is tracked.
+        framework (`str`, *optional*, defaults to `"pytorch"`): The framework for the example.
+    """
+    if is_offline_mode():
+        return
+
+    data = {"example": example_name, "framework": framework}
+    for args in example_args:
+        args_as_dict = {k: v for k, v in args.__dict__.items() if not k.startswith("_") and v is not None}
+        if "model_name_or_path" in args_as_dict:
+            model_name = args_as_dict["model_name_or_path"]
+            # Filter out local paths
+            if not os.path.isdir(model_name):
+                data["model_name"] = args_as_dict["model_name_or_path"]
+        if "dataset_name" in args_as_dict:
+            data["dataset_name"] = args_as_dict["dataset_name"]
+        elif "task_name" in args_as_dict:
+            # Extract script name from the example_name
+            script_name = example_name.replace("tf_", "").replace("flax_", "").replace("run_", "")
+            script_name = script_name.replace("_no_trainer", "")
+            data["dataset_name"] = f"{script_name}-{args_as_dict['task_name']}"
+
+    headers = {"user-agent": http_user_agent(data)}
+    try:
+        r = requests.head(HUGGINGFACE_CO_EXAMPLES_TELEMETRY, headers=headers)
+        r.raise_for_status()
+    except Exception:
+        # We don't want to error in case of connection errors of any kind.
+        pass

templates/adding_a_new_example_script/{{cookiecutter.directory_name}}/run_{{cookiecutter.example_shortcut}}.py

@@ -46,6 +46,7 @@ from transformers import (
     set_seed,
 )
 from transformers.trainer_utils import get_last_checkpoint
+from transformers.utils import send_example_telemetry
 
 
 logger = logging.getLogger(__name__)
@@ -207,6 +208,10 @@ def main():
     else:
         model_args, data_args, training_args = parser.parse_args_into_dataclasses()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_{{cookiecutter.example_shortcut}}", model_args, data_args)
+
     # Detecting last checkpoint.
     last_checkpoint = None
     if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
@@ -519,6 +524,7 @@ from transformers import (
     get_scheduler,
     set_seed,
 )
+from transformers.utils import send_example_telemetry
 
 
 logger = logging.getLogger(__name__)
@@ -662,6 +668,10 @@ def parse_args():
 def main():
     args = parse_args()
 
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_{{cookiecutter.example_shortcut}", args)
+
     # Initialize the accelerator. We will let the accelerator handle device placement for us in this example.
     accelerator = Accelerator()
     # Make one log on every process with the configuration for debugging.