[None][feat] Add the PrimTS native sparse MLA FMHA for DeepSeek-V4 generation and enable it by default on SM100 - #20209

[None][feat] Add the PrimTS native sparse MLA FMHA for DeepSeek-V4 generation and enable it by default on SM100#20209

yunruis merged 19 commits into

NVIDIA:mainNVIDIA/TensorRT-LLM:mainfrom

yunruis:user/yunruis/dsv4-prims-ts-sparse-mlayunruis/TensorRT-LLM:user/yunruis/dsv4-prims-ts-sparse-mlaCopy head branch name to clipboard