CoWA, a structured attention architecture that distributes access to the causal history across KV heads, shows that full causal coverage can be a collective property of the head ensemble rather than a duplicated property of every head.
Jing-Ze Shi, Zhang-Yang Peng, Xian-Duo Li et al.· 0 citations
KernelGenBench is presented, the first unified multi-source and multi-chip infrastructure for evaluating LLM- and agent-generated Triton kernels and establishes operator source, hardware platform, and agentic scaffold as distinct dimensions of kernel-generation capability, and shows that success in a familiar source-ha...
Pei-Yu Zang, Jian-Hang Tao, Jia-Ling Zhang et al.· 0 citations
It is shown that agent-based methods consistently outperform pure LLM sampling methods, while cuBLAS operators are the most challenging across all methods, and a unified benchmark for systematically evaluating LLM-and agent-generated Triton kernels across diverse operator sources and heterogeneous hardware platforms is...