Skip to content

Commit cede942

Browse files
b8zhongb8zhong
authored
[Benchmark] Add support for multiple batch size benchmark through CLI in benchmark_moe.py (vllm-project#20516)
Signed-off-by: Brayden Zhong <b8zhong@uwaterloo.ca>
1 parent fe1e924 commit cede942

File tree

2 files changed

+149
-2
lines changed

2 files changed

+149
-2
lines changed

‎benchmarks/kernels/benchmark_moe.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -620,7 +620,7 @@ def main(args: argparse.Namespace):
620620
4096,
621621
]
622622
else:
623-
batch_sizes = [args.batch_size]
623+
batch_sizes = args.batch_size
624624

625625
use_deep_gemm = bool(args.use_deep_gemm)
626626

@@ -728,7 +728,7 @@ def _distribute(method: str, inputs: list[Any]) -> list[Any]:
728728
)
729729
parser.add_argument("--use-deep-gemm", action="store_true")
730730
parser.add_argument("--seed", type=int, default=0)
731-
parser.add_argument("--batch-size", type=int, required=False)
731+
parser.add_argument("--batch-size", type=int, nargs="+", required=False)
732732
parser.add_argument("--tune", action="store_true")
733733
parser.add_argument("--trust-remote-code", action="store_true")
734734
parser.add_argument("--model-prefix", type=str, required=False)

‎vllm/model_executor/layers/fused_moe/configs/E=16,N=1024,device_name=NVIDIA_B200,dtype=fp8_w8a8.json

Lines changed: 147 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,147 @@
1+
{
2+
"1": {
3+
"BLOCK_SIZE_M": 16,
4+
"BLOCK_SIZE_N": 32,
5+
"BLOCK_SIZE_K": 256,
6+
"GROUP_SIZE_M": 64,
7+
"num_warps": 4,
8+
"num_stages": 4
9+
},
10+
"2": {
11+
"BLOCK_SIZE_M": 16,
12+
"BLOCK_SIZE_N": 32,
13+
"BLOCK_SIZE_K": 256,
14+
"GROUP_SIZE_M": 1,
15+
"num_warps": 4,
16+
"num_stages": 3
17+
},
18+
"4": {
19+
"BLOCK_SIZE_M": 16,
20+
"BLOCK_SIZE_N": 32,
21+
"BLOCK_SIZE_K": 256,
22+
"GROUP_SIZE_M": 1,
23+
"num_warps": 4,
24+
"num_stages": 3
25+
},
26+
"8": {
27+
"BLOCK_SIZE_M": 16,
28+
"BLOCK_SIZE_N": 128,
29+
"BLOCK_SIZE_K": 128,
30+
"GROUP_SIZE_M": 16,
31+
"num_warps": 4,
32+
"num_stages": 5
33+
},
34+
"16": {
35+
"BLOCK_SIZE_M": 16,
36+
"BLOCK_SIZE_N": 64,
37+
"BLOCK_SIZE_K": 128,
38+
"GROUP_SIZE_M": 1,
39+
"num_warps": 4,
40+
"num_stages": 4
41+
},
42+
"24": {
43+
"BLOCK_SIZE_M": 16,
44+
"BLOCK_SIZE_N": 128,
45+
"BLOCK_SIZE_K": 128,
46+
"GROUP_SIZE_M": 1,
47+
"num_warps": 4,
48+
"num_stages": 4
49+
},
50+
"32": {
51+
"BLOCK_SIZE_M": 16,
52+
"BLOCK_SIZE_N": 64,
53+
"BLOCK_SIZE_K": 128,
54+
"GROUP_SIZE_M": 1,
55+
"num_warps": 4,
56+
"num_stages": 5
57+
},
58+
"48": {
59+
"BLOCK_SIZE_M": 16,
60+
"BLOCK_SIZE_N": 64,
61+
"BLOCK_SIZE_K": 128,
62+
"GROUP_SIZE_M": 1,
63+
"num_warps": 4,
64+
"num_stages": 5
65+
},
66+
"64": {
67+
"BLOCK_SIZE_M": 16,
68+
"BLOCK_SIZE_N": 128,
69+
"BLOCK_SIZE_K": 256,
70+
"GROUP_SIZE_M": 1,
71+
"num_warps": 8,
72+
"num_stages": 3
73+
},
74+
"96": {
75+
"BLOCK_SIZE_M": 16,
76+
"BLOCK_SIZE_N": 128,
77+
"BLOCK_SIZE_K": 256,
78+
"GROUP_SIZE_M": 1,
79+
"num_warps": 8,
80+
"num_stages": 3
81+
},
82+
"128": {
83+
"BLOCK_SIZE_M": 16,
84+
"BLOCK_SIZE_N": 128,
85+
"BLOCK_SIZE_K": 256,
86+
"GROUP_SIZE_M": 1,
87+
"num_warps": 8,
88+
"num_stages": 3
89+
},
90+
"256": {
91+
"BLOCK_SIZE_M": 16,
92+
"BLOCK_SIZE_N": 128,
93+
"BLOCK_SIZE_K": 128,
94+
"GROUP_SIZE_M": 1,
95+
"num_warps": 4,
96+
"num_stages": 4
97+
},
98+
"512": {
99+
"BLOCK_SIZE_M": 64,
100+
"BLOCK_SIZE_N": 128,
101+
"BLOCK_SIZE_K": 128,
102+
"GROUP_SIZE_M": 1,
103+
"num_warps": 8,
104+
"num_stages": 3
105+
},
106+
"1024": {
107+
"BLOCK_SIZE_M": 64,
108+
"BLOCK_SIZE_N": 128,
109+
"BLOCK_SIZE_K": 128,
110+
"GROUP_SIZE_M": 1,
111+
"num_warps": 4,
112+
"num_stages": 3
113+
},
114+
"1536": {
115+
"BLOCK_SIZE_M": 128,
116+
"BLOCK_SIZE_N": 128,
117+
"BLOCK_SIZE_K": 64,
118+
"GROUP_SIZE_M": 1,
119+
"num_warps": 4,
120+
"num_stages": 4
121+
},
122+
"2048": {
123+
"BLOCK_SIZE_M": 128,
124+
"BLOCK_SIZE_N": 128,
125+
"BLOCK_SIZE_K": 64,
126+
"GROUP_SIZE_M": 1,
127+
"num_warps": 4,
128+
"num_stages": 4
129+
},
130+
"3072": {
131+
"BLOCK_SIZE_M": 256,
132+
"BLOCK_SIZE_N": 256,
133+
"BLOCK_SIZE_K": 128,
134+
"GROUP_SIZE_M": 1,
135+
"num_warps": 8,
136+
"num_stages": 3
137+
},
138+
"4096": {
139+
"BLOCK_SIZE_M": 128,
140+
"BLOCK_SIZE_N": 256,
141+
"BLOCK_SIZE_K": 128,
142+
"GROUP_SIZE_M": 1,
143+
"num_warps": 8,
144+
"num_stages": 4
145+
}
146+
}
147+

0 commit comments

Comments
 (0)