obscura/vllm
2
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2026-06-06 00:16:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

[Perf] Add tuned selective_state_update configs for H200 and RTX PRO … (#44251)

Browse Source 
Signed-off-by: Majid Taheri Andani <tahemaji@amazon.com>
Co-authored-by: Majid Taheri Andani <tahemaji@amazon.com>
Co-authored-by: tomeras91 <57313761+tomeras91@users.noreply.github.com>
This commit is contained in:
Majid
2026-06-02 23:59:01 -07:00
committed by GitHub
parent 87954eb50e
commit 9af53a3c13
4 changed files with 348 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vllm/model_executor/layers/mamba/ops/configs/selective_state_update/headdim=64,dstate=128,device_name=NVIDIA_H200,cache_dtype=float16.json
+87
View File
@@ -0,0 +1,87 @@
{
"triton_version": "3.6.0",
"8": {
"BLOCK_SIZE_M": 4,
"num_warps": 2
},
"16": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"32": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"64": {
"BLOCK_SIZE_M": 16,
"num_warps": 4
},
"128": {
"BLOCK_SIZE_M": 8,
"num_warps": 2
},
"256": {
"BLOCK_SIZE_M": 8,
"num_warps": 2
},
"512": {
"BLOCK_SIZE_M": 16,
"num_warps": 1
},
"1024": {
"BLOCK_SIZE_M": 16,
"num_warps": 1
},
"2048": {
"BLOCK_SIZE_M": 8,
"num_warps": 2
},
"4096": {
"BLOCK_SIZE_M": 16,
"num_warps": 2
},
"8192": {
"BLOCK_SIZE_M": 32,
"num_warps": 2
},
"12288": {
"BLOCK_SIZE_M": 32,
"num_warps": 4
},
"16384": {
"BLOCK_SIZE_M": 16,
"num_warps": 2
},
"24576": {
"BLOCK_SIZE_M": 32,
"num_warps": 4
},
"32768": {
"BLOCK_SIZE_M": 32,
"num_warps": 2
},
"49152": {
"BLOCK_SIZE_M": 16,
"num_warps": 2
},
"65536": {
"BLOCK_SIZE_M": 16,
"num_warps": 2
},
"98304": {
"BLOCK_SIZE_M": 16,
"num_warps": 2
},
"131072": {
"BLOCK_SIZE_M": 16,
"num_warps": 2
},
"196608": {
"BLOCK_SIZE_M": 16,
"num_warps": 2
},
"262144": {
"BLOCK_SIZE_M": 16,
"num_warps": 2
}
}
vllm/model_executor/layers/mamba/ops/configs/selective_state_update/headdim=64,dstate=128,device_name=NVIDIA_H200,cache_dtype=float32.json
+87
View File
@@ -0,0 +1,87 @@
{
"triton_version": "3.6.0",
"8": {
"BLOCK_SIZE_M": 8,
"num_warps": 4
},
"16": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"32": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"64": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"128": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"256": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"512": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"1024": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"2048": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"4096": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"8192": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"12288": {
"BLOCK_SIZE_M": 8,
"num_warps": 1
},
"16384": {
"BLOCK_SIZE_M": 8,
"num_warps": 1
},
"24576": {
"BLOCK_SIZE_M": 8,
"num_warps": 1
},
"32768": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"49152": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"65536": {
"BLOCK_SIZE_M": 8,
"num_warps": 2
},
"98304": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"131072": {
"BLOCK_SIZE_M": 32,
"num_warps": 4
},
"196608": {
"BLOCK_SIZE_M": 16,
"num_warps": 1
},
"262144": {
"BLOCK_SIZE_M": 16,
"num_warps": 1
}
}
vllm/model_executor/layers/mamba/ops/configs/selective_state_update/headdim=64,dstate=128,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Server_Edition,cache_dtype=float16.json
+87
View File
@@ -0,0 +1,87 @@
{
"triton_version": "3.6.0",
"8": {
"BLOCK_SIZE_M": 4,
"num_warps": 4
},
"16": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"32": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"64": {
"BLOCK_SIZE_M": 8,
"num_warps": 8
},
"128": {
"BLOCK_SIZE_M": 16,
"num_warps": 8
},
"256": {
"BLOCK_SIZE_M": 16,
"num_warps": 8
},
"512": {
"BLOCK_SIZE_M": 16,
"num_warps": 8
},
"1024": {
"BLOCK_SIZE_M": 16,
"num_warps": 8
},
"2048": {
"BLOCK_SIZE_M": 16,
"num_warps": 8
},
"4096": {
"BLOCK_SIZE_M": 16,
"num_warps": 8
},
"8192": {
"BLOCK_SIZE_M": 32,
"num_warps": 1
},
"12288": {
"BLOCK_SIZE_M": 16,
"num_warps": 1
},
"16384": {
"BLOCK_SIZE_M": 32,
"num_warps": 4
},
"24576": {
"BLOCK_SIZE_M": 32,
"num_warps": 4
},
"32768": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"49152": {
"BLOCK_SIZE_M": 32,
"num_warps": 2
},
"65536": {
"BLOCK_SIZE_M": 32,
"num_warps": 1
},
"98304": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"131072": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"196608": {
"BLOCK_SIZE_M": 32,
"num_warps": 1
},
"262144": {
"BLOCK_SIZE_M": 16,
"num_warps": 1
}
}
vllm/model_executor/layers/mamba/ops/configs/selective_state_update/headdim=64,dstate=128,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Server_Edition,cache_dtype=float32.json
+87
View File
@@ -0,0 +1,87 @@
{
"triton_version": "3.6.0",
"8": {
"BLOCK_SIZE_M": 8,
"num_warps": 8
},
"16": {
"BLOCK_SIZE_M": 8,
"num_warps": 8
},
"32": {
"BLOCK_SIZE_M": 8,
"num_warps": 8
},
"64": {
"BLOCK_SIZE_M": 8,
"num_warps": 8
},
"128": {
"BLOCK_SIZE_M": 8,
"num_warps": 8
},
"256": {
"BLOCK_SIZE_M": 8,
"num_warps": 8
},
"512": {
"BLOCK_SIZE_M": 8,
"num_warps": 8
},
"1024": {
"BLOCK_SIZE_M": 16,
"num_warps": 8
},
"2048": {
"BLOCK_SIZE_M": 16,
"num_warps": 8
},
"4096": {
"BLOCK_SIZE_M": 16,
"num_warps": 1
},
"8192": {
"BLOCK_SIZE_M": 4,
"num_warps": 8
},
"12288": {
"BLOCK_SIZE_M": 16,
"num_warps": 1
},
"16384": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"24576": {
"BLOCK_SIZE_M": 4,
"num_warps": 1
},
"32768": {
"BLOCK_SIZE_M": 4,
"num_warps": 4
},
"49152": {
"BLOCK_SIZE_M": 16,
"num_warps": 4
},
"65536": {
"BLOCK_SIZE_M": 64,
"num_warps": 8
},
"98304": {
"BLOCK_SIZE_M": 16,
"num_warps": 1
},
"131072": {
"BLOCK_SIZE_M": 8,
"num_warps": 1
},
"196608": {
"BLOCK_SIZE_M": 64,
"num_warps": 8
},
"262144": {
"BLOCK_SIZE_M": 64,
"num_warps": 4
}
}