[None][feat] Add Mamba2 MTP SSM cache CUDA kernel for tree-based speculative decoding#12537

JadoTu merged 4 commits intoNVIDIA:mainNVIDIA/TensorRT-LLM:mainfrom

JadoTu:mamba2_tree_based_mtp_CUDA_kernelJadoTu/TensorRT-LLM:mamba2_tree_based_mtp_CUDA_kernelCopy head branch name to clipboard