v0.2.6

What's Changed

• fix: choose cuda arthitectures based on cuda version by @guocuimi in #463
• kernel: add grouped gemm support for moe by @guocuimi in #458
• kernel: added oob handling for grouped gemm kernel by @guocuimi in #465
• refactor: add _1 into stride for contiguous dim by @guocuimi in #466
• ci: set cuda arch to native for ci workflows by @guocuimi in #467
• refactor: move TileShape into launch_mha_kernel_sm80 by @guocuimi in #468
• refactor: split attention kernel into collective mainloop, collective epilogue and kernel by @guocuimi in #469
• fix: skip failed unittests for blackwell gpus by @guocuimi in #472
• feat: added single tile scheduler for attn kernel by @guocuimi in #473
• feat: add tile scheduler for grouped gemm and refactor gemm kernel by @guocuimi in #474
• refactor: split mla kernels into collective_mla collective_epilogue by @guocuimi in #475
• feat: use global residue_mnk for oob handling by @guocuimi in #476
• feat: simplify mask logic to avoid manual index computation by @guocuimi in #477
• feat: added static permistent tile scheduler with swizzle and rasterize by @guocuimi in #478
• feat: added gtest_main with filters based on compute_capabilities by @guocuimi in #479
• ci: upgrade cutlass to v4.1 and switch to forked repo by @guocuimi in #481
• feat: add tma copy for paged kv by @guocuimi in #480
• feat: added gather tma copy to control smem box size by @guocuimi in #482
• feat: use aggressive compress-mode for fatbin by @guocuimi in #484
• feat: added fast StaticPersistentTileScheduler for 1d tma multicast by @guocuimi in #485
• feat: [1/n] added sm120 fmha using collective async copy by @guocuimi in #483
• feat: [2/n] added warp specialization kernel for sm120 fmha by @guocuimi in #486
• refactor: move kernel code into different folders by @guocuimi in #487
• feat: added KV multi-stages support for attn sm120 by @guocuimi in #489
• refactor: simplify mha block tiling logic by @guocuimi in #488
• feat: added smem and gmem layout selector for attn kernel by @guocuimi in #490
• feat: added args and params for attn kernels by @guocuimi in #491
• feat: added universal fmha runner by @guocuimi in #492
• feat: added kernel builder for attn by @guocuimi in #493
• refactor: change stride for Q/K/V to MNKL by @guocuimi in #494
• upgrade torch to 12.8 by @guocuimi in #496
• ci: fix nccl related build error by @guocuimi in #497

Full Changelog: v0.2.5...v0.2.6