From c8c2f838e7d9f1975ee00ac206c9cbbe331a2f86 Mon Sep 17 00:00:00 2001 From: vanshbhatia-amd Date: Tue, 7 Jul 2026 18:19:08 -0400 Subject: [PATCH] Add tuned selective_state_update config for AMD Instinct MI355 (#47767) Signed-off-by: vanshbhatia-amd <210711135+vanshbhatia-amd@users.noreply.github.com> Co-authored-by: vanshbhatia-amd <210711135+vanshbhatia-amd@users.noreply.github.com> --- ...nstinct_MI355_OAM,cache_dtype=float16.json | 35 +++++++++++++++++++ 1 file changed, 35 insertions(+) create mode 100644 vllm/model_executor/layers/mamba/ops/configs/selective_state_update/headdim=64,dstate=128,device_name=AMD_Instinct_MI355_OAM,cache_dtype=float16.json diff --git a/vllm/model_executor/layers/mamba/ops/configs/selective_state_update/headdim=64,dstate=128,device_name=AMD_Instinct_MI355_OAM,cache_dtype=float16.json b/vllm/model_executor/layers/mamba/ops/configs/selective_state_update/headdim=64,dstate=128,device_name=AMD_Instinct_MI355_OAM,cache_dtype=float16.json new file mode 100644 index 0000000000..8a78ec1a4b --- /dev/null +++ b/vllm/model_executor/layers/mamba/ops/configs/selective_state_update/headdim=64,dstate=128,device_name=AMD_Instinct_MI355_OAM,cache_dtype=float16.json @@ -0,0 +1,35 @@ +{ + "triton_version": "3.6.0", + "10": { + "BLOCK_SIZE_M": 16, + "num_warps": 4 + }, + "80": { + "BLOCK_SIZE_M": 16, + "num_warps": 4 + }, + "160": { + "BLOCK_SIZE_M": 16, + "num_warps": 4 + }, + "320": { + "BLOCK_SIZE_M": 16, + "num_warps": 4 + }, + "640": { + "BLOCK_SIZE_M": 32, + "num_warps": 4 + }, + "1280": { + "BLOCK_SIZE_M": 16, + "num_warps": 1 + }, + "2560": { + "BLOCK_SIZE_M": 16, + "num_warps": 1 + }, + "5120": { + "BLOCK_SIZE_M": 16, + "num_warps": 1 + } +} \ No newline at end of file