TritonAscendBench

Gxj230958

A migration benchmark that measures whether production GPU Triton kernels — extracted from vLLM and SGLang inference paths — can be ported to the Huawei Ascend 910B (Atlas A2) via `triton-ascend` without losing correctness or performance.