The method identifies that FP4 errors stem from low-entropy tokens and uses trace-aligned training to reach 65.94% AIME-120 accuracy on R1-Qwen-14B.
Rebellions AI offers switchless Dragonfly and 6D hypercube topologies for Rebel 100s clusters, reducing capital and power costs while maintaining competitive performance for large-scale GenAI inference.
The Rebel100s accelerator integrates switch functionality into I/O chiplets to support up to 256 sockets without external switches.
Rebellions presented its AI inference chip strategy at RAISE Summit 2026.
Wert Intelligence combined Rebellions' NPU with PlutoLM to build a domestic AI full-stack solution for patent data classification and summarization.
Rebellions ATOM NPU servers doubled the inference efficiency of a UAE water purification robot.
At the Seoul meetup, participants confirmed the trend of vLLM growing into standard infrastructure for LLM serving.
A 4-chiplet AI SoC implements a full-chip mesh with 16Gbps UCIe-Advanced.
Mongolia Customs automated HS Code classification with an LLM/RAG-based NPU chatbot.
SK Telecom's X Caliber will drop GPUs for ATOM™ NPU in 2025.
Rebellions and SqueezeBits held a hands-on workshop where participants ran vLLM directly on NPUs.
NPUs and VIB ESS dramatically cut AI data center power costs and peak loads.
This is an event page reporting on the first vLLM meetup held in Korea.
At Hot Chips 2025, REBEL-Quad's world's first UCIe-Advanced NPU demonstration was unveiled on-site.
IGLOO implemented the on-premise LLM security assistant AiR using Rebellions NPUs.
Rebellions unveiled a high-efficiency LLM serving technology compatible with vLLM by optimizing FlashAttention and other techniques for NPUs.
KOLON BENIT has launched an AI vision safety management solution for construction sites that maximizes power efficiency using Rebellions' NPU.
AI Briefing