Here is a simple sequence classification pipeline using MLFlow, which supports (1) model training; (2) model evaluation; (3) model registering on Azure Machine Learning.
Submit MLFlow experiment job to Azure ML from local path
pip install -U azureml-mlflow
cd ./azure/
python submit.py \
--workspace_name=<workspace-name> \
--subscription_id=<subscription-id> \
--resource_group=<resource-group> \
--experiment_name=<experiment-name> \
--compute_target=<compute-target> \
--num_gpus=1 \
--datastore_name=<datastore-name> \
--model_name_or_path=bert-base-uncased \
--train_blob_path=<train-blob-path> \
--train_file=train.json \
--dev_blob_path=<dev-blob-path> \
--dev_file=dev.json \
--test_blob_path=<test-blob-path> \
--test_file=test.json \
--max_seq_length=128 \
--batch_size=16 \
--gradient_accumulation_steps=1 \
--learning_rate=3e-05 \
--num_epochs=3.0 \
--weight_decay=0.01 \
--warmup_steps=1000 \
--eval_steps=500 \
--output_dir=./output/ \
--seed=42 \
--save_total_limit=10 \
--registered_name=<registered-model-name>Track job progress and review results on Azure ML
Figure 8: the details of training
Figure 9: the classification report for evaluation
Figure 10: the model registry