Skip to content

Commit b126da4

Browse files
committed
fp8 tuned configs-v1 | mixtral8x(7B, 22B) TP=1,2,4,8
1 parent 820c7aa commit b126da4

8 files changed

+1952
-0
lines changed
Lines changed: 244 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,244 @@
1+
{
2+
"1": {
3+
"BLOCK_SIZE_M": 16,
4+
"BLOCK_SIZE_N": 64,
5+
"BLOCK_SIZE_K": 256,
6+
"GROUP_SIZE_M": 1,
7+
"num_warps": 4,
8+
"num_stages": 2,
9+
"waves_per_eu": 0,
10+
"matrix_instr_nonkdim": 16,
11+
"kpack": 2
12+
},
13+
"2": {
14+
"BLOCK_SIZE_M": 16,
15+
"BLOCK_SIZE_N": 64,
16+
"BLOCK_SIZE_K": 256,
17+
"GROUP_SIZE_M": 1,
18+
"num_warps": 4,
19+
"num_stages": 2,
20+
"waves_per_eu": 0,
21+
"matrix_instr_nonkdim": 16,
22+
"kpack": 2
23+
},
24+
"4": {
25+
"BLOCK_SIZE_M": 16,
26+
"BLOCK_SIZE_N": 64,
27+
"BLOCK_SIZE_K": 256,
28+
"GROUP_SIZE_M": 1,
29+
"num_warps": 4,
30+
"num_stages": 2,
31+
"waves_per_eu": 0,
32+
"matrix_instr_nonkdim": 16,
33+
"kpack": 2
34+
},
35+
"8": {
36+
"BLOCK_SIZE_M": 16,
37+
"BLOCK_SIZE_N": 64,
38+
"BLOCK_SIZE_K": 256,
39+
"GROUP_SIZE_M": 1,
40+
"num_warps": 4,
41+
"num_stages": 2,
42+
"waves_per_eu": 0,
43+
"matrix_instr_nonkdim": 16,
44+
"kpack": 2
45+
},
46+
"16": {
47+
"BLOCK_SIZE_M": 16,
48+
"BLOCK_SIZE_N": 64,
49+
"BLOCK_SIZE_K": 256,
50+
"GROUP_SIZE_M": 1,
51+
"num_warps": 4,
52+
"num_stages": 2,
53+
"waves_per_eu": 0,
54+
"matrix_instr_nonkdim": 16,
55+
"kpack": 2
56+
},
57+
"24": {
58+
"BLOCK_SIZE_M": 16,
59+
"BLOCK_SIZE_N": 64,
60+
"BLOCK_SIZE_K": 256,
61+
"GROUP_SIZE_M": 1,
62+
"num_warps": 4,
63+
"num_stages": 2,
64+
"waves_per_eu": 0,
65+
"matrix_instr_nonkdim": 16,
66+
"kpack": 2
67+
},
68+
"32": {
69+
"BLOCK_SIZE_M": 16,
70+
"BLOCK_SIZE_N": 64,
71+
"BLOCK_SIZE_K": 256,
72+
"GROUP_SIZE_M": 1,
73+
"num_warps": 4,
74+
"num_stages": 2,
75+
"waves_per_eu": 0,
76+
"matrix_instr_nonkdim": 16,
77+
"kpack": 2
78+
},
79+
"48": {
80+
"BLOCK_SIZE_M": 16,
81+
"BLOCK_SIZE_N": 64,
82+
"BLOCK_SIZE_K": 256,
83+
"GROUP_SIZE_M": 1,
84+
"num_warps": 4,
85+
"num_stages": 2,
86+
"waves_per_eu": 0,
87+
"matrix_instr_nonkdim": 16,
88+
"kpack": 2
89+
},
90+
"64": {
91+
"BLOCK_SIZE_M": 32,
92+
"BLOCK_SIZE_N": 64,
93+
"BLOCK_SIZE_K": 256,
94+
"GROUP_SIZE_M": 1,
95+
"num_warps": 4,
96+
"num_stages": 2,
97+
"waves_per_eu": 0,
98+
"matrix_instr_nonkdim": 16,
99+
"kpack": 2
100+
},
101+
"96": {
102+
"BLOCK_SIZE_M": 32,
103+
"BLOCK_SIZE_N": 64,
104+
"BLOCK_SIZE_K": 256,
105+
"GROUP_SIZE_M": 1,
106+
"num_warps": 4,
107+
"num_stages": 2,
108+
"waves_per_eu": 0,
109+
"matrix_instr_nonkdim": 16,
110+
"kpack": 2
111+
},
112+
"128": {
113+
"BLOCK_SIZE_M": 32,
114+
"BLOCK_SIZE_N": 64,
115+
"BLOCK_SIZE_K": 256,
116+
"GROUP_SIZE_M": 1,
117+
"num_warps": 4,
118+
"num_stages": 2,
119+
"waves_per_eu": 0,
120+
"matrix_instr_nonkdim": 16,
121+
"kpack": 2
122+
},
123+
"256": {
124+
"BLOCK_SIZE_M": 32,
125+
"BLOCK_SIZE_N": 64,
126+
"BLOCK_SIZE_K": 256,
127+
"GROUP_SIZE_M": 1,
128+
"num_warps": 4,
129+
"num_stages": 2,
130+
"waves_per_eu": 0,
131+
"matrix_instr_nonkdim": 16,
132+
"kpack": 2
133+
},
134+
"512": {
135+
"BLOCK_SIZE_M": 32,
136+
"BLOCK_SIZE_N": 64,
137+
"BLOCK_SIZE_K": 256,
138+
"GROUP_SIZE_M": 1,
139+
"num_warps": 4,
140+
"num_stages": 2,
141+
"waves_per_eu": 0,
142+
"matrix_instr_nonkdim": 16,
143+
"kpack": 2
144+
},
145+
"1024": {
146+
"BLOCK_SIZE_M": 64,
147+
"BLOCK_SIZE_N": 64,
148+
"BLOCK_SIZE_K": 256,
149+
"GROUP_SIZE_M": 1,
150+
"num_warps": 4,
151+
"num_stages": 2,
152+
"waves_per_eu": 0,
153+
"matrix_instr_nonkdim": 16,
154+
"kpack": 2
155+
},
156+
"1536": {
157+
"BLOCK_SIZE_M": 64,
158+
"BLOCK_SIZE_N": 64,
159+
"BLOCK_SIZE_K": 256,
160+
"GROUP_SIZE_M": 1,
161+
"num_warps": 4,
162+
"num_stages": 2,
163+
"waves_per_eu": 0,
164+
"matrix_instr_nonkdim": 16,
165+
"kpack": 2
166+
},
167+
"2048": {
168+
"BLOCK_SIZE_M": 64,
169+
"BLOCK_SIZE_N": 64,
170+
"BLOCK_SIZE_K": 256,
171+
"GROUP_SIZE_M": 1,
172+
"num_warps": 4,
173+
"num_stages": 2,
174+
"waves_per_eu": 0,
175+
"matrix_instr_nonkdim": 16,
176+
"kpack": 2
177+
},
178+
"3072": {
179+
"BLOCK_SIZE_M": 64,
180+
"BLOCK_SIZE_N": 64,
181+
"BLOCK_SIZE_K": 256,
182+
"GROUP_SIZE_M": 1,
183+
"num_warps": 4,
184+
"num_stages": 2,
185+
"waves_per_eu": 0,
186+
"matrix_instr_nonkdim": 16,
187+
"kpack": 2
188+
},
189+
"4096": {
190+
"BLOCK_SIZE_M": 64,
191+
"BLOCK_SIZE_N": 64,
192+
"BLOCK_SIZE_K": 256,
193+
"GROUP_SIZE_M": 1,
194+
"num_warps": 4,
195+
"num_stages": 2,
196+
"waves_per_eu": 0,
197+
"matrix_instr_nonkdim": 16,
198+
"kpack": 2
199+
},
200+
"8192": {
201+
"BLOCK_SIZE_M": 64,
202+
"BLOCK_SIZE_N": 64,
203+
"BLOCK_SIZE_K": 256,
204+
"GROUP_SIZE_M": 1,
205+
"num_warps": 4,
206+
"num_stages": 2,
207+
"waves_per_eu": 0,
208+
"matrix_instr_nonkdim": 16,
209+
"kpack": 2
210+
},
211+
"16384": {
212+
"BLOCK_SIZE_M": 64,
213+
"BLOCK_SIZE_N": 64,
214+
"BLOCK_SIZE_K": 256,
215+
"GROUP_SIZE_M": 1,
216+
"num_warps": 4,
217+
"num_stages": 2,
218+
"waves_per_eu": 0,
219+
"matrix_instr_nonkdim": 16,
220+
"kpack": 2
221+
},
222+
"18432": {
223+
"BLOCK_SIZE_M": 64,
224+
"BLOCK_SIZE_N": 64,
225+
"BLOCK_SIZE_K": 256,
226+
"GROUP_SIZE_M": 1,
227+
"num_warps": 4,
228+
"num_stages": 2,
229+
"waves_per_eu": 0,
230+
"matrix_instr_nonkdim": 16,
231+
"kpack": 2
232+
},
233+
"20480": {
234+
"BLOCK_SIZE_M": 64,
235+
"BLOCK_SIZE_N": 64,
236+
"BLOCK_SIZE_K": 256,
237+
"GROUP_SIZE_M": 1,
238+
"num_warps": 4,
239+
"num_stages": 2,
240+
"waves_per_eu": 0,
241+
"matrix_instr_nonkdim": 16,
242+
"kpack": 2
243+
}
244+
}

0 commit comments

Comments
 (0)