Skip to content

Commit a018e55

Browse files
Ximingwang-09ximing.wxm
and
ximing.wxm
authored
[Kernel] Add fp8_w8a8 fused MoE kernel tuning configs for DeepSeek V3/R1 on NVIDIA H20 (vllm-project#16753)
Signed-off-by: ximing.wxm <[email protected]> Co-authored-by: ximing.wxm <[email protected]>
1 parent 6211b92 commit a018e55

File tree

1 file changed

+146
-0
lines changed

1 file changed

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

0 commit comments

Comments
 (0)