Thanks for sending an requirement! Please fill in the following template to help quickly solve your problem.
1.mxfp4&mxfp8输入 fp32输出WeightND全切tensorapi；2.通过tensorapi 新增支持mxfp4 mxfp8输入weightnz及bf16输出、fp32输出各类dtype组合：3.gmmfr fp32&bf16输出性能优化
Thanks for sending an requirement! Please fill in the following template to help quickly solve your problem.
Backgroud（背景信息）
1.mxfp4&mxfp8输入 fp32输出WeightND全切tensorapi；2.通过tensorapi 新增支持mxfp4 mxfp8输入weightnz及bf16输出、fp32输出各类dtype组合：3.gmmfr fp32&bf16输出性能优化
Benefit / Necessity （价值/作用）
Design（设计方案）