matx
MatX is an AI-hardware company designing specialized semiconductors optimized for large language model workloads. Its flagship MatX One chip targets high-throughput, low-latency training, inference, and reinforcement learning for large dense and Mixture-of-Experts models, storing weights in SRAM to cut latency and reaching over 2,000 output tokens…
原文来源AI 开源 Releases
查看官方原文
MatX is an AI-hardware company designing specialized semiconductors optimized for large language model workloads. Its flagship MatX One chip targets high-throughput, low-latency training, inference, and reinforcement learning for large dense and Mixture-of-Experts models, storing weights in SRAM to cut latency and reaching over 2,000 output tokens…