Union

About Union

Union key features

• **Model Training & Fine-Tuning**: Supports large-scale training and fine-tuning on GPU clusters with features for reproducibility, efficiency, and collaboration.
• **Data Processing**: Enables the creation of complex data processing pipelines with built-in operators, dynamic resource allocation, and advanced error handling.
• **Inference**: Provides high-throughput, reliable, and fault-tolerant inference workflows that can dynamically scale resources based on workload demands.
• **GenAI & LLMs**: Facilitates the deployment of generative AI applications with features for building reproducible RAG data pipelines and optimizing batch inference.
• **Bioinformatics & Synthetic Biology**: Offers tools for scientific computing workflows, including dependency management, data flow integration, and accelerated dataset handling.

Union use cases

• Model Training & Fine-Tuning: Enable large-scale training or fine-tuning of machine learning models using GPU clusters across various environments.
• Data Processing: Build and manage complex data processing pipelines that are reliable, scalable, and optimized for performance.
• Inference: Deliver high-throughput and fault-tolerant inference workflows for production-scale AI applications, dynamically scaling resources as needed.
• GenAI & LLMs: Create and optimize generative AI applications, including building reproducible RAG data pipelines and parallelizing LLM batch inference.
• Bioinformatics & Synthetic Biology: Conduct scientific computing workflows, including generating novel DNA sequences and predicting protein structures, with strict versioning and data flow management.

Useful for

• **Efficient Collaboration**: Union Serverless facilitates collaboration between engineering, operations, and data science teams, enabling them to create AI products more efficiently and with reduced costs.
• **Optimized Performance**: The platform offers optimized performance for running complex AI workloads, ensuring faster execution times and efficient resource utilization.
• **Scalable Infrastructure**: Union supports dynamic scaling of resources, allowing users to adjust resource allocation based on workload demands, which helps minimize costs while maintaining performance.
• **Robust Workflow Management**: It provides a unified platform for orchestrating the entire machine learning lifecycle, from data processing to model training and inference, ensuring consistency and traceability.
• **Enhanced Reliability**: The system includes fault-tolerant execution and error recovery features, ensuring that workflows can automatically recover from failures without manual intervention.

Price

• **Free Plan**: DIY ML orchestration for teams with on-premise or bare metal deployments; open source (Apache 2.0); Union support plans available.
• **Serverless Plan**: Pay-as-you-go with $30 in free credit; optimized and expanded Flyte; limited to 1 seat; pay only for consumed resources.
• **BYOC – Startup Plan**: $500/month plus a percentage of compute; ideal for small teams; up to 5 seats; single cluster/cloud; standard bring-your-own-cloud deployment.
• **BYOC – Enterprise Plan**: Custom pricing with committed use discounts; built for large-scale deployments; unlimited seats; multi-cluster/cloud; customizable bring-your-own-cloud deployment.
• **Serverless Compute Pricing**: CPU at $0.12 per core-hour; GPU T4 at $0.71 per GPU-hour; GPU A100 (40GB) at $4.48 per GPU-hour; memory at $0.029 per GB-hour.