ASTRA: Communication-Efficient Acceleration for Multi-Device Transformer Inference 文章

ArXiv CS.AI2026-05-28NEWSen作者: Xiao Liu, Lijun Zhang, Deepak Ganesan, Hui Guan

查看原文 →

ASTRA: Communication-Efficient Acceleration for Multi-Device Transformer Inference · 相关技术

相关技术

transformers Distributed Class Tokens Noise-Augmented Quantization mixed-precision attention Sparse Sequence Parallelism