GitLab

issue/8: causalsoftmax算子-昇腾

Issue/48 Rope CPU & CUDA

使用maskedInplaceOp和SoftmaxOp实现CausalSoftmaxOp

Issue/127/feat. General Elementwise Framework with Refactored SwiGLU (CPU & CUDA)

issue/127: change meta within ElementwiseInfo to std::vector<size_t> for correct alignment and change the reference name of the Opaque struct to Opaque instead of struct Opaque

issue/127: Add arguments to CREATE_ELEMENTWISE_PLATFORM_DESCRIPTOR macros for indirecting variable names, change DeviceImpl to use Result for the return type of the create function, change CEIL_DIV

issue/127: Optimize elementwise CUDA code by removing redundancy, change/correct kernel logic when all inputs have the same dtype

issue/127: Refactor ElementwiseInfo, refactor elementwise to use workspace for storing meta, fix misc. issues

issue/127: fix CUDA mix-precision broadcasting input mismatch issue, adjust comment structure and template variable order

enable_if, remove std::move() in elementwise_cpu.h, add <array> inclusion

issue/127: refactor ElementwiseInfo to use utils::Result, change elementwise calcualte and calculateImpl to return infiniStatus_t, add CHECK_CUDA to cuda function calls

issue/127: modify swiglu test to correctly handle broadcast scenarios, add two broadcast testcases, correct elementwise cpu mix-precision implementation

issue/127: refactor elementwise framework, complete CUDA implementation, refactor swiglu using the generic elementwise framework

issue/40: 实现沐曦rms_norm算子

issue/130: GEMM算子CPU平台的omp重构

issue/32: 实现摩尔线程rms_norm算子

issue/111: 添加rmsnorm以及算子编译流程

issue/4 添加cuda causal softmax算子