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=float32.json b/vllm/model_executor/layers/mamba/ops/configs/selective_state_update/headdim=64,dstate=128,device_name=AMD_Instinct_MI355_OAM,cache_dtype=float32.json index d9f535e320a..a23f123eefe 100644 --- a/vllm/model_executor/layers/mamba/ops/configs/selective_state_update/headdim=64,dstate=128,device_name=AMD_Instinct_MI355_OAM,cache_dtype=float32.json +++ b/vllm/model_executor/layers/mamba/ops/configs/selective_state_update/headdim=64,dstate=128,device_name=AMD_Instinct_MI355_OAM,cache_dtype=float32.json @@ -1,35 +1,51 @@ { "triton_version": "3.6.0", - "10": { + "128": { + "BLOCK_SIZE_M": 16, + "num_warps": 4 + }, + "256": { + "BLOCK_SIZE_M": 16, + "num_warps": 4 + }, + "1024": { "BLOCK_SIZE_M": 8, - "num_warps": 4 + "num_warps": 1 }, - "80": { + "2048": { "BLOCK_SIZE_M": 8, - "num_warps": 4 + "num_warps": 1 }, - "160": { + "4096": { "BLOCK_SIZE_M": 16, - "num_warps": 4 + "num_warps": 8 }, - "320": { - "BLOCK_SIZE_M": 16, - "num_warps": 4 - }, - "640": { - "BLOCK_SIZE_M": 16, - "num_warps": 4 - }, - "1280": { - "BLOCK_SIZE_M": 16, + "8192": { + "BLOCK_SIZE_M": 8, "num_warps": 2 }, - "2560": { - "BLOCK_SIZE_M": 32, + "16384": { + "BLOCK_SIZE_M": 8, + "num_warps": 2 + }, + "32768": { + "BLOCK_SIZE_M": 64, "num_warps": 4 }, - "5120": { - "BLOCK_SIZE_M": 32, - "num_warps": 4 + "65536": { + "BLOCK_SIZE_M": 64, + "num_warps": 2 + }, + "131072": { + "BLOCK_SIZE_M": 64, + "num_warps": 2 + }, + "196608": { + "BLOCK_SIZE_M": 64, + "num_warps": 2 + }, + "262144": { + "BLOCK_SIZE_M": 64, + "num_warps": 2 } } \ No newline at end of file