邁爾凌MLSteam深度學習訓練解決方案

To generate a production grade DNN model, a developer will need to go through many difficult and time consuming steps, including dataset collection, dataset cleansing, dataset labeling, dataset augmentation, dataset format conversion, models selection, model design, hyperparameters tuning, model training, model evaluation, and model format conversion. Each step requires different tools and configurations that require time and effort for preparation, and switching between these tools often requires additional time writing code to convert different formats to use with different tools.

GIGABYTE’s DNN Training Appliance aims to reduce this complexity by providing a complete training and management platform, while incorporating all these processes into an easy to use web-browser based GUI. Users can import, convert and manage their dataset; design, train and evaluate different DNN models; and test inferencing of their trained models. Based on GIGABYTE’s G481-HA1 server, the platform is fully optimized to use the bare metal resources available to deliver improved training performance on cost-efficient hardware.

DNN models need to be trained on a large dataset to achieve an acceptable level of accuracy. Depending on the dataset size, this training could take days or even weeks. And in order to adapt to the latest business circumstances or situations (such as new products, new regulations, etc.), the DNN model needs to be periodically retrained through the latest datasets. If running a DNN training job takes too long, it will have a serious impact on an organization’s operations, resource management, and competiveness.

GIGABYTE’s DNN Training Appliance helps to reduce training time by incorporating many different optimization features:GPU memory optimization to accommodate a large amount of training input or to fit a large model into GPU memory, automatic hyperparameter tuning (during a training job) to achieve higher accuracy, and dataset cleaning features to reduce the training time generated by mislabeled or duplicated training data.

Cloud IDE & Utilities Interface
Users can easily create a Cloud IDE (based on Jupyterlab) for DNN model development or data preprocessing by attaching their dataset. The Cloud IDE also provides utilities, such as hyperparameter passing, 3rd-party IDE integration (VSCode and PyCharm), tensorboard and GPU monitoring to simplify the training process.

Model Templates and Optimization Tutorial
Gigabyte’s DNN Training sytem has built-in templates, guiding the user on how to train different types of models (for image classification, object detection, etc.) with various optimization techniques, such as GPU memory optimization and mixed precision training. These templates allow the user to easily choose the dataset, DNN models, and hyperparameter settings needed based on the DNN application type. Thus, the user can easily leverage templates for collaboration.

Dataset Annotation Visualization
The platform supports multiple dataset annotated formats so that the user can preview the annotated dataset on the dataset page. Ex. Bounding box, segmentation images, etc.

System Resource Monitoring
GIGABYTE’s DNN Training Appliance features real-time GPU (including GPU utilization, GPU memory usage, and temperature), CPU, Disk, and memory usage monitoring.

Administration Dashboard
GIGABYTE’s DNN Training Appliance features a dashboard for administration, including an audit log, training tasks overview, dataset overview, and user account management.