Platform Features
Model Life Cycle Management
Platform support AI model life cycle management for different stages like data gathering and preparation, model training, model evaluation, model deployment, and model governance. This enables seamless exploration and deployment of multiple AI models.
Model Repository
Platform supports a built-in model repository to store input and output artifacts of models like training data, model metrics, and actual ML models. A model repository makes these artifacts accessible in a distributed environment across nodes in a cluster.
Data Pipelines
High-performance data pipelines for data ingestion, transformation, and processing. Pipeline variants can be batch, stream, or real-time, with wide-ranging protocols like SCP, S/FTP, HTTP/S, MQTT, AQMP, etc. supported.
Model Store
Platform support extensive model store including
• Supervised Models - Supervised and Supervised Unbalanced Models
• Reinforcement Models: RL and Deep RL Models.
• LLM Models: Vision Models , NLP /Text Models, Document AI Models.
• Other Models: Anomaly Models, Time Series Models , Causal Models.
Resource Management
Resource allocation, monitoring and auditing for model training and inference tasks. Platform integrates with industry leading workload managers (e.g . slurm) for managing hardware resources like CPU, GPU, memory etc.
GPU Acceleration
Platform has built in support of using NVIDIA CUDA and cuDNN Software for distributed training and NVIDIA TensorRT for inference activities, enabling applications to utilize available GPU infrastructure.