GitLab

issue/1050 - fix paged caching and paged prefill on metax

issue/1048: 修复QY机器上的paged_attention_prefill和swiglu编译问题

issue/1035 - support both int32 and int64 in kv caching

issue/1042 - fix address misaligned error of rearrange

issue/1035: 添加NVIDIA平台上的kv_caching算子

issue/1041 - feat: use template to replace int64_t in paged_attention_prefill kernel with test pass

issue/1032 - fix block size on iluvatar

Issue/1032n - cuda swiglu

Issue/1036 paged caching support strides

Issue/1021： issue/1021 - feat: support bf16 in infiniccl with mccl

issue/1025 - temporarily disable paged op compilation on hygon

Demo-131 Cuda graph with optimized paged attention

issue/961: fix metax init with preload

Issue/1008