Rebellions SDK

Deploy with Confidence from Day One.

Contact Sales

Built for developers who scale AI

Rebellions delivers the performance and efficiency of custom AI silicon—without compromising the software experience.
From PyTorch and vLLM to Triton, our stack offers a seamless, GPU-like workflow optimized for real-world AI inference at scale.

Unified Inference Stack That Just Works

A GPU-class software experience designed for scalable AI inference

  • Native integration with PyTorch, vLLM, and Triton
  • Broad model support including MoE and decoder-only LLMs
  • Production-proven in high-concurrency, latency-sensitiev environments

PyTorch-Native Inference Engine

Purpose-built for real-world PyTorch workflows

  • Graph-mode optimization for hardware-accelerated execution
  • Eager mode for flexible experimentation and prototyping
  • Distributed inference via Collective Communication Library
  • Intelligent precision control across FP32, FP16, FP8, FP6, and FP4

vLLM Serving, Optimized for Concurrency

High-throughput LLM serving with native PyTorch support

  • Ideal for high QPS workloads with long-context models
  • Standard PyTorch API compatibility
  • Hugging Face integration for fast model onboarding

Triton SDK & Developer Tooling

Low-level programmability meets developer-friendly design

  • Triton backend with custom kernel programming support
  • Intuitive APIs for seamless integration into ML pipelines
  • Built-in profiling and debugging tools
  • Extensive docs and real-world example codes