Empowering Innovation Through
Open Source
At ScitiX, we open our tools and technologies to the world, fostering collaboration and driving global progress in AI infrastructure.
Available Resource
SiPipe: Revolutionizing LLM Inference with CPU-GPU Synergy
SiPipe is a cutting-edge inference engine that bridges the CPU–GPU utilization gap in pipeline-parallel large language model (LLM) deployments. By intelligently offloading auxiliary computations to underutilized CPU resources, SiPipe enhances throughput and reduces latency without compromising GPU performance. Its innovative design incorporates CPU sampling, a token-safe execution model, and structure-aware transmission to eliminate execution bubbles and optimize resource utilization. In extensive testing across various LLMs, SiPipe demonstrated up to 2.1× higher throughput, 43% lower per-token latency, and up to 23% improved GPU utilization compared to the state-of-the-art vLLM under identical pipeline-parallel configurations.
Arks: Effortless LLM at Scale
Arks is a cloud-native framework that simplifies deploying, orchestrating, and scaling LLM-based applications on Kubernetes. It handles inference workloads efficiently, manages resources across CPU and GPU, and integrates seamlessly with existing workflows. By removing operational complexity, Arks lets teams focus on building AI-driven applications while maximizing performance, reliability, and scalability in production.
Sichek: Proactive Node Health
Sichek is a smart tool that detects and diagnoses node health issues before they impact performance. It provides real-time visibility into potential hardware failures or bottlenecks, helping upstream systems like Kubernetes or task managers respond proactively. By enabling early intervention and timely task rescheduling, Sichek ensures high GPU utilization, smooth operations, and maximum efficiency across your AI infrastructure.
Self-service
With self-service ordering and a comprehensive user guide to walk you through every step, the docs have you covered whether you're prototyping or scaling.
Open Source
Our infrastructure is built in the open - transparent, verifiable, and always
evolving with the community.
Explore our GitHub to see what we're building, contribute your ideas, or just star
the repo to stay connected.
Contact Us
