A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 18:30 UTC
MI325 configs, fused_moe_kernel bugfix (#14987)
Signed-off-by: Eugene Kuznetsov <eugene.kuznetsov@amd.com>
ekuznetsov139 committed
Mar 18, 2025 at 08:05 UTC
8b793f7ec639afed42c7814d4f97134f311108e8
54 files changed
+8878
-1
vllm/model_executor/layers/fused_moe/configs/E=256,N=1024,device_name=AMD_Instinct_MI325X,block_shape=[128,128].json
new
+200
@@ -0,0 +1,200 @@
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": 128,
16
+ "BLOCK_SIZE_K": 64,
17
+ "GROUP_SIZE_M": 1,
18
+ "num_warps": 8,
19
+ "num_stages": 2,
20
+ "waves_per_eu": 0,
21
+ "matrix_instr_nonkdim": 16,
22
+ "kpack": 1
23
+ },
24
+ "4": {
25
+ "BLOCK_SIZE_M": 16,
26
+ "BLOCK_SIZE_N": 128,
27
+ "BLOCK_SIZE_K": 128,
28
+ "GROUP_SIZE_M": 1,
29
+ "num_warps": 8,
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": 128,
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": 128,
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": 128,
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": 16,
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": 16,
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": 16,
92
+ "BLOCK_SIZE_N": 128,
93
+ "BLOCK_SIZE_K": 128,
94
+ "GROUP_SIZE_M": 32,
95
+ "num_warps": 8,
96
+ "num_stages": 2,
97
+ "waves_per_eu": 0,
98
+ "matrix_instr_nonkdim": 16,
99
+ "kpack": 1
100
+ },
101
+ "96": {
102
+ "BLOCK_SIZE_M": 16,
103
+ "BLOCK_SIZE_N": 64,
104
+ "BLOCK_SIZE_K": 128,
105
+ "GROUP_SIZE_M": 32,
106
+ "num_warps": 1,
107
+ "num_stages": 2,
108
+ "waves_per_eu": 0,
109
+ "matrix_instr_nonkdim": 16,
110
+ "kpack": 2
111
+ },
112
+ "128": {
113
+ "BLOCK_SIZE_M": 16,
114
+ "BLOCK_SIZE_N": 64,
115
+ "BLOCK_SIZE_K": 128,
116
+ "GROUP_SIZE_M": 32,
117
+ "num_warps": 1,
118
+ "num_stages": 2,
119
+ "waves_per_eu": 0,
120
+ "matrix_instr_nonkdim": 16,
121
+ "kpack": 2
122
+ },
123
+ "256": {
124
+ "BLOCK_SIZE_M": 64,
125
+ "BLOCK_SIZE_N": 64,
126
+ "BLOCK_SIZE_K": 128,
127
+ "GROUP_SIZE_M": 32,
128
+ "num_warps": 8,
129
+ "num_stages": 2,
130
+ "waves_per_eu": 0,
131
+ "matrix_instr_nonkdim": 16,
132
+ "kpack": 1
133
+ },
134
+ "512": {
135
+ "BLOCK_SIZE_M": 64,
136
+ "BLOCK_SIZE_N": 64,
137
+ "BLOCK_SIZE_K": 128,
138
+ "GROUP_SIZE_M": 8,
139
+ "num_warps": 8,
140
+ "num_stages": 2,
141
+ "waves_per_eu": 0,
142
+ "matrix_instr_nonkdim": 16,
143
+ "kpack": 1
144
+ },
145
+ "1024": {
146
+ "BLOCK_SIZE_M": 64,
147
+ "BLOCK_SIZE_N": 64,
148
+ "BLOCK_SIZE_K": 128,
149
+ "GROUP_SIZE_M": 8,
150
+ "num_warps": 8,
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": 128,
159
+ "BLOCK_SIZE_K": 128,
160
+ "GROUP_SIZE_M": 8,
161
+ "num_warps": 8,
162
+ "num_stages": 2,
163
+ "waves_per_eu": 0,
164
+ "matrix_instr_nonkdim": 16,
165
+ "kpack": 2
166
+ },
167
+ "2048": {
168
+ "BLOCK_SIZE_M": 128,
169
+ "BLOCK_SIZE_N": 128,
170
+ "BLOCK_SIZE_K": 64,
171
+ "GROUP_SIZE_M": 16,
172
+ "num_warps": 8,
173
+ "num_stages": 2,
174
+ "waves_per_eu": 0,
175
+ "matrix_instr_nonkdim": 16,
176
+ "kpack": 2
177
+ },
178
+ "3072": {
179
+ "BLOCK_SIZE_M": 128,
180
+ "BLOCK_SIZE_N": 128,
181
+ "BLOCK_SIZE_K": 64,
182
+ "GROUP_SIZE_M": 8,
183
+ "num_warps": 8,
184
+ "num_stages": 2,
185
+ "waves_per_eu": 0,
186
+ "matrix_instr_nonkdim": 16,
187
+ "kpack": 2
188
+ },
189
+ "4096": {
190
+ "BLOCK_SIZE_M": 128,
191
+ "BLOCK_SIZE_N": 128,
192
+ "BLOCK_SIZE_K": 64,
193
+ "GROUP_SIZE_M": 8,
194
+ "num_warps": 8,
195
+ "num_stages": 2,
196
+ "waves_per_eu": 0,
197
+ "matrix_instr_nonkdim": 16,
198
+ "kpack": 2
199
+ }
200
+}
vllm/model_executor/layers/fused_moe/configs/E=256,N=1024,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+200
@@ -0,0 +1,200 @@
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": 128,
16
+ "BLOCK_SIZE_K": 64,
17
+ "GROUP_SIZE_M": 1,
18
+ "num_warps": 8,
19
+ "num_stages": 2,
20
+ "waves_per_eu": 0,
21
+ "matrix_instr_nonkdim": 16,
22
+ "kpack": 1
23
+ },
24
+ "4": {
25
+ "BLOCK_SIZE_M": 16,
26
+ "BLOCK_SIZE_N": 128,
27
+ "BLOCK_SIZE_K": 128,
28
+ "GROUP_SIZE_M": 1,
29
+ "num_warps": 8,
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": 128,
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": 128,
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": 128,
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": 16,
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": 16,
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": 16,
92
+ "BLOCK_SIZE_N": 128,
93
+ "BLOCK_SIZE_K": 128,
94
+ "GROUP_SIZE_M": 32,
95
+ "num_warps": 8,
96
+ "num_stages": 2,
97
+ "waves_per_eu": 0,
98
+ "matrix_instr_nonkdim": 16,
99
+ "kpack": 1
100
+ },
101
+ "96": {
102
+ "BLOCK_SIZE_M": 16,
103
+ "BLOCK_SIZE_N": 64,
104
+ "BLOCK_SIZE_K": 128,
105
+ "GROUP_SIZE_M": 32,
106
+ "num_warps": 1,
107
+ "num_stages": 2,
108
+ "waves_per_eu": 0,
109
+ "matrix_instr_nonkdim": 16,
110
+ "kpack": 2
111
+ },
112
+ "128": {
113
+ "BLOCK_SIZE_M": 16,
114
+ "BLOCK_SIZE_N": 64,
115
+ "BLOCK_SIZE_K": 128,
116
+ "GROUP_SIZE_M": 32,
117
+ "num_warps": 1,
118
+ "num_stages": 2,
119
+ "waves_per_eu": 0,
120
+ "matrix_instr_nonkdim": 16,
121
+ "kpack": 2
122
+ },
123
+ "256": {
124
+ "BLOCK_SIZE_M": 64,
125
+ "BLOCK_SIZE_N": 64,
126
+ "BLOCK_SIZE_K": 128,
127
+ "GROUP_SIZE_M": 32,
128
+ "num_warps": 8,
129
+ "num_stages": 2,
130
+ "waves_per_eu": 0,
131
+ "matrix_instr_nonkdim": 16,
132
+ "kpack": 1
133
+ },
134
+ "512": {
135
+ "BLOCK_SIZE_M": 64,
136
+ "BLOCK_SIZE_N": 64,
137
+ "BLOCK_SIZE_K": 128,
138
+ "GROUP_SIZE_M": 8,
139
+ "num_warps": 8,
140
+ "num_stages": 2,
141
+ "waves_per_eu": 0,
142
+ "matrix_instr_nonkdim": 16,
143
+ "kpack": 1
144
+ },
145
+ "1024": {
146
+ "BLOCK_SIZE_M": 64,
147
+ "BLOCK_SIZE_N": 64,
148
+ "BLOCK_SIZE_K": 128,
149
+ "GROUP_SIZE_M": 8,
150
+ "num_warps": 8,
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": 128,
159
+ "BLOCK_SIZE_K": 128,
160
+ "GROUP_SIZE_M": 8,
161
+ "num_warps": 8,
162
+ "num_stages": 2,
163
+ "waves_per_eu": 0,
164
+ "matrix_instr_nonkdim": 16,
165
+ "kpack": 2
166
+ },
167
+ "2048": {
168
+ "BLOCK_SIZE_M": 128,
169
+ "BLOCK_SIZE_N": 128,
170
+ "BLOCK_SIZE_K": 64,
171
+ "GROUP_SIZE_M": 16,
172
+ "num_warps": 8,
173
+ "num_stages": 2,
174
+ "waves_per_eu": 0,
175
+ "matrix_instr_nonkdim": 16,
176
+ "kpack": 2
177
+ },
178
+ "3072": {
179
+ "BLOCK_SIZE_M": 128,
180
+ "BLOCK_SIZE_N": 128,
181
+ "BLOCK_SIZE_K": 64,
182
+ "GROUP_SIZE_M": 8,
183
+ "num_warps": 8,
184
+ "num_stages": 2,
185
+ "waves_per_eu": 0,
186
+ "matrix_instr_nonkdim": 16,
187
+ "kpack": 2
188
+ },
189
+ "4096": {
190
+ "BLOCK_SIZE_M": 128,
191
+ "BLOCK_SIZE_N": 128,
192
+ "BLOCK_SIZE_K": 64,
193
+ "GROUP_SIZE_M": 8,
194
+ "num_warps": 8,
195
+ "num_stages": 2,
196
+ "waves_per_eu": 0,
197
+ "matrix_instr_nonkdim": 16,
198
+ "kpack": 2
199
+ }
200
+}
vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+200
@@ -0,0 +1,200 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_M": 16,
4
+ "BLOCK_SIZE_N": 16,
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": 1
12
+ },
13
+ "2": {
14
+ "BLOCK_SIZE_M": 16,
15
+ "BLOCK_SIZE_N": 64,
16
+ "BLOCK_SIZE_K": 128,
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": 1
23
+ },
24
+ "4": {
25
+ "BLOCK_SIZE_M": 64,
26
+ "BLOCK_SIZE_N": 64,
27
+ "BLOCK_SIZE_K": 128,
28
+ "GROUP_SIZE_M": 1,
29
+ "num_warps": 8,
30
+ "num_stages": 2,
31
+ "waves_per_eu": 0,
32
+ "matrix_instr_nonkdim": 16,
33
+ "kpack": 1
34
+ },
35
+ "8": {
36
+ "BLOCK_SIZE_M": 16,
37
+ "BLOCK_SIZE_N": 64,
38
+ "BLOCK_SIZE_K": 128,
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": 1
45
+ },
46
+ "16": {
47
+ "BLOCK_SIZE_M": 16,
48
+ "BLOCK_SIZE_N": 64,
49
+ "BLOCK_SIZE_K": 128,
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": 1
56
+ },
57
+ "24": {
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 128,
60
+ "BLOCK_SIZE_K": 64,
61
+ "GROUP_SIZE_M": 1,
62
+ "num_warps": 8,
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": 128,
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": 128,
83
+ "GROUP_SIZE_M": 1,
84
+ "num_warps": 2,
85
+ "num_stages": 2,
86
+ "waves_per_eu": 0,
87
+ "matrix_instr_nonkdim": 16,
88
+ "kpack": 2
89
+ },
90
+ "64": {
91
+ "BLOCK_SIZE_M": 16,
92
+ "BLOCK_SIZE_N": 64,
93
+ "BLOCK_SIZE_K": 128,
94
+ "GROUP_SIZE_M": 1,
95
+ "num_warps": 2,
96
+ "num_stages": 2,
97
+ "waves_per_eu": 0,
98
+ "matrix_instr_nonkdim": 16,
99
+ "kpack": 2
100
+ },
101
+ "96": {
102
+ "BLOCK_SIZE_M": 16,
103
+ "BLOCK_SIZE_N": 32,
104
+ "BLOCK_SIZE_K": 128,
105
+ "GROUP_SIZE_M": 4,
106
+ "num_warps": 2,
107
+ "num_stages": 2,
108
+ "waves_per_eu": 0,
109
+ "matrix_instr_nonkdim": 16,
110
+ "kpack": 2
111
+ },
112
+ "128": {
113
+ "BLOCK_SIZE_M": 16,
114
+ "BLOCK_SIZE_N": 32,
115
+ "BLOCK_SIZE_K": 128,
116
+ "GROUP_SIZE_M": 4,
117
+ "num_warps": 1,
118
+ "num_stages": 2,
119
+ "waves_per_eu": 0,
120
+ "matrix_instr_nonkdim": 16,
121
+ "kpack": 2
122
+ },
123
+ "256": {
124
+ "BLOCK_SIZE_M": 16,
125
+ "BLOCK_SIZE_N": 32,
126
+ "BLOCK_SIZE_K": 128,
127
+ "GROUP_SIZE_M": 1,
128
+ "num_warps": 2,
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": 128,
138
+ "GROUP_SIZE_M": 8,
139
+ "num_warps": 8,
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": 128,
148
+ "BLOCK_SIZE_K": 128,
149
+ "GROUP_SIZE_M": 8,
150
+ "num_warps": 8,
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": 128,
159
+ "BLOCK_SIZE_K": 128,
160
+ "GROUP_SIZE_M": 8,
161
+ "num_warps": 8,
162
+ "num_stages": 2,
163
+ "waves_per_eu": 0,
164
+ "matrix_instr_nonkdim": 16,
165
+ "kpack": 2
166
+ },
167
+ "2048": {
168
+ "BLOCK_SIZE_M": 128,
169
+ "BLOCK_SIZE_N": 128,
170
+ "BLOCK_SIZE_K": 64,
171
+ "GROUP_SIZE_M": 8,
172
+ "num_warps": 8,
173
+ "num_stages": 2,
174
+ "waves_per_eu": 0,
175
+ "matrix_instr_nonkdim": 16,
176
+ "kpack": 2
177
+ },
178
+ "3072": {
179
+ "BLOCK_SIZE_M": 128,
180
+ "BLOCK_SIZE_N": 128,
181
+ "BLOCK_SIZE_K": 64,
182
+ "GROUP_SIZE_M": 8,
183
+ "num_warps": 8,
184
+ "num_stages": 2,
185
+ "waves_per_eu": 0,
186
+ "matrix_instr_nonkdim": 16,
187
+ "kpack": 2
188
+ },
189
+ "4096": {
190
+ "BLOCK_SIZE_M": 128,
191
+ "BLOCK_SIZE_N": 128,
192
+ "BLOCK_SIZE_K": 64,
193
+ "GROUP_SIZE_M": 8,
194
+ "num_warps": 8,
195
+ "num_stages": 2,
196
+ "waves_per_eu": 0,
197
+ "matrix_instr_nonkdim": 16,
198
+ "kpack": 2
199
+ }
200
+}
vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+200
@@ -0,0 +1,200 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_M": 16,
4
+ "BLOCK_SIZE_N": 16,
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": 1
12
+ },
13
+ "2": {
14
+ "BLOCK_SIZE_M": 16,
15
+ "BLOCK_SIZE_N": 64,
16
+ "BLOCK_SIZE_K": 128,
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": 1
23
+ },
24
+ "4": {
25
+ "BLOCK_SIZE_M": 64,
26
+ "BLOCK_SIZE_N": 64,
27
+ "BLOCK_SIZE_K": 128,
28
+ "GROUP_SIZE_M": 1,
29
+ "num_warps": 8,
30
+ "num_stages": 2,
31
+ "waves_per_eu": 0,
32
+ "matrix_instr_nonkdim": 16,
33
+ "kpack": 1
34
+ },
35
+ "8": {
36
+ "BLOCK_SIZE_M": 16,
37
+ "BLOCK_SIZE_N": 64,
38
+ "BLOCK_SIZE_K": 128,
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": 1
45
+ },
46
+ "16": {
47
+ "BLOCK_SIZE_M": 16,
48
+ "BLOCK_SIZE_N": 64,
49
+ "BLOCK_SIZE_K": 128,
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": 1
56
+ },
57
+ "24": {
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 128,
60
+ "BLOCK_SIZE_K": 64,
61
+ "GROUP_SIZE_M": 1,
62
+ "num_warps": 8,
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": 128,
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": 128,
83
+ "GROUP_SIZE_M": 1,
84
+ "num_warps": 2,
85
+ "num_stages": 2,
86
+ "waves_per_eu": 0,
87
+ "matrix_instr_nonkdim": 16,
88
+ "kpack": 2
89
+ },
90
+ "64": {
91
+ "BLOCK_SIZE_M": 16,
92
+ "BLOCK_SIZE_N": 64,
93
+ "BLOCK_SIZE_K": 128,
94
+ "GROUP_SIZE_M": 1,
95
+ "num_warps": 2,
96
+ "num_stages": 2,
97
+ "waves_per_eu": 0,
98
+ "matrix_instr_nonkdim": 16,
99
+ "kpack": 2
100
+ },
101
+ "96": {
102
+ "BLOCK_SIZE_M": 16,
103
+ "BLOCK_SIZE_N": 32,
104
+ "BLOCK_SIZE_K": 128,
105
+ "GROUP_SIZE_M": 4,
106
+ "num_warps": 2,
107
+ "num_stages": 2,
108
+ "waves_per_eu": 0,
109
+ "matrix_instr_nonkdim": 16,
110
+ "kpack": 2
111
+ },
112
+ "128": {
113
+ "BLOCK_SIZE_M": 16,
114
+ "BLOCK_SIZE_N": 32,
115
+ "BLOCK_SIZE_K": 128,
116
+ "GROUP_SIZE_M": 4,
117
+ "num_warps": 1,
118
+ "num_stages": 2,
119
+ "waves_per_eu": 0,
120
+ "matrix_instr_nonkdim": 16,
121
+ "kpack": 2
122
+ },
123
+ "256": {
124
+ "BLOCK_SIZE_M": 16,
125
+ "BLOCK_SIZE_N": 32,
126
+ "BLOCK_SIZE_K": 128,
127
+ "GROUP_SIZE_M": 1,
128
+ "num_warps": 2,
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": 128,
138
+ "GROUP_SIZE_M": 8,
139
+ "num_warps": 8,
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": 128,
148
+ "BLOCK_SIZE_K": 128,
149
+ "GROUP_SIZE_M": 8,
150
+ "num_warps": 8,
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": 128,
159
+ "BLOCK_SIZE_K": 128,
160
+ "GROUP_SIZE_M": 8,
161
+ "num_warps": 8,
162
+ "num_stages": 2,
163
+ "waves_per_eu": 0,
164
+ "matrix_instr_nonkdim": 16,
165
+ "kpack": 2
166
+ },
167
+ "2048": {
168
+ "BLOCK_SIZE_M": 128,
169
+ "BLOCK_SIZE_N": 128,
170
+ "BLOCK_SIZE_K": 64,
171
+ "GROUP_SIZE_M": 8,
172
+ "num_warps": 8,
173
+ "num_stages": 2,
174
+ "waves_per_eu": 0,
175
+ "matrix_instr_nonkdim": 16,
176
+ "kpack": 2
177
+ },
178
+ "3072": {
179
+ "BLOCK_SIZE_M": 128,
180
+ "BLOCK_SIZE_N": 128,
181
+ "BLOCK_SIZE_K": 64,
182
+ "GROUP_SIZE_M": 8,
183
+ "num_warps": 8,
184
+ "num_stages": 2,
185
+ "waves_per_eu": 0,
186
+ "matrix_instr_nonkdim": 16,
187
+ "kpack": 2
188
+ },
189
+ "4096": {
190
+ "BLOCK_SIZE_M": 128,
191
+ "BLOCK_SIZE_N": 128,
192
+ "BLOCK_SIZE_K": 64,
193
+ "GROUP_SIZE_M": 8,
194
+ "num_warps": 8,
195
+ "num_stages": 2,
196
+ "waves_per_eu": 0,
197
+ "matrix_instr_nonkdim": 16,
198
+ "kpack": 2
199
+ }
200
+}
vllm/model_executor/layers/fused_moe/configs/E=256,N=512,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+200
@@ -0,0 +1,200 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_M": 16,
4
+ "BLOCK_SIZE_N": 16,
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": 1
12
+ },
13
+ "2": {
14
+ "BLOCK_SIZE_M": 16,
15
+ "BLOCK_SIZE_N": 64,
16
+ "BLOCK_SIZE_K": 128,
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": 1
23
+ },
24
+ "4": {
25
+ "BLOCK_SIZE_M": 64,
26
+ "BLOCK_SIZE_N": 64,
27
+ "BLOCK_SIZE_K": 128,
28
+ "GROUP_SIZE_M": 1,
29
+ "num_warps": 8,
30
+ "num_stages": 2,
31
+ "waves_per_eu": 0,
32
+ "matrix_instr_nonkdim": 16,
33
+ "kpack": 1
34
+ },
35
+ "8": {
36
+ "BLOCK_SIZE_M": 16,
37
+ "BLOCK_SIZE_N": 64,
38
+ "BLOCK_SIZE_K": 128,
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": 1
45
+ },
46
+ "16": {
47
+ "BLOCK_SIZE_M": 16,
48
+ "BLOCK_SIZE_N": 64,
49
+ "BLOCK_SIZE_K": 128,
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": 1
56
+ },
57
+ "24": {
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 128,
60
+ "BLOCK_SIZE_K": 64,
61
+ "GROUP_SIZE_M": 1,
62
+ "num_warps": 8,
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": 128,
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": 128,
83
+ "GROUP_SIZE_M": 1,
84
+ "num_warps": 2,
85
+ "num_stages": 2,
86
+ "waves_per_eu": 0,
87
+ "matrix_instr_nonkdim": 16,
88
+ "kpack": 2
89
+ },
90
+ "64": {
91
+ "BLOCK_SIZE_M": 16,
92
+ "BLOCK_SIZE_N": 64,
93
+ "BLOCK_SIZE_K": 128,
94
+ "GROUP_SIZE_M": 1,
95
+ "num_warps": 2,
96
+ "num_stages": 2,
97
+ "waves_per_eu": 0,
98
+ "matrix_instr_nonkdim": 16,
99
+ "kpack": 2
100
+ },
101
+ "96": {
102
+ "BLOCK_SIZE_M": 16,
103
+ "BLOCK_SIZE_N": 32,
104
+ "BLOCK_SIZE_K": 128,
105
+ "GROUP_SIZE_M": 4,
106
+ "num_warps": 2,
107
+ "num_stages": 2,
108
+ "waves_per_eu": 0,
109
+ "matrix_instr_nonkdim": 16,
110
+ "kpack": 2
111
+ },
112
+ "128": {
113
+ "BLOCK_SIZE_M": 16,
114
+ "BLOCK_SIZE_N": 32,
115
+ "BLOCK_SIZE_K": 128,
116
+ "GROUP_SIZE_M": 4,
117
+ "num_warps": 1,
118
+ "num_stages": 2,
119
+ "waves_per_eu": 0,
120
+ "matrix_instr_nonkdim": 16,
121
+ "kpack": 2
122
+ },
123
+ "256": {
124
+ "BLOCK_SIZE_M": 16,
125
+ "BLOCK_SIZE_N": 32,
126
+ "BLOCK_SIZE_K": 128,
127
+ "GROUP_SIZE_M": 1,
128
+ "num_warps": 2,
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": 128,
138
+ "GROUP_SIZE_M": 8,
139
+ "num_warps": 8,
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": 128,
148
+ "BLOCK_SIZE_K": 128,
149
+ "GROUP_SIZE_M": 8,
150
+ "num_warps": 8,
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": 128,
159
+ "BLOCK_SIZE_K": 128,
160
+ "GROUP_SIZE_M": 8,
161
+ "num_warps": 8,
162
+ "num_stages": 2,
163
+ "waves_per_eu": 0,
164
+ "matrix_instr_nonkdim": 16,
165
+ "kpack": 2
166
+ },
167
+ "2048": {
168
+ "BLOCK_SIZE_M": 128,
169
+ "BLOCK_SIZE_N": 128,
170
+ "BLOCK_SIZE_K": 64,
171
+ "GROUP_SIZE_M": 8,
172
+ "num_warps": 8,
173
+ "num_stages": 2,
174
+ "waves_per_eu": 0,
175
+ "matrix_instr_nonkdim": 16,
176
+ "kpack": 2
177
+ },
178
+ "3072": {
179
+ "BLOCK_SIZE_M": 128,
180
+ "BLOCK_SIZE_N": 128,
181
+ "BLOCK_SIZE_K": 64,
182
+ "GROUP_SIZE_M": 8,
183
+ "num_warps": 8,
184
+ "num_stages": 2,
185
+ "waves_per_eu": 0,
186
+ "matrix_instr_nonkdim": 16,
187
+ "kpack": 2
188
+ },
189
+ "4096": {
190
+ "BLOCK_SIZE_M": 128,
191
+ "BLOCK_SIZE_N": 128,
192
+ "BLOCK_SIZE_K": 64,
193
+ "GROUP_SIZE_M": 8,
194
+ "num_warps": 8,
195
+ "num_stages": 2,
196
+ "waves_per_eu": 0,
197
+ "matrix_instr_nonkdim": 16,
198
+ "kpack": 2
199
+ }
200
+}
vllm/model_executor/layers/fused_moe/fused_moe.py
+6
-1
@@ -783,8 +783,12 @@ def invoke_fused_moe_kernel(A: torch.Tensor,
783
use_int8_w8a16=use_int8_w8a16,
784
**config,
785
)
786
-
786
else:
787
+ config = config.copy()
788
+ BLOCK_SIZE_K = config.pop("BLOCK_SIZE_K")
789
+ if block_shape is not None:
790
+ BLOCK_SIZE_K = min(BLOCK_SIZE_K, min(block_shape[0],
791
+ block_shape[1]))
792
fused_moe_kernel[grid](
793
A,
794
B,
@@ -823,6 +827,7 @@ def invoke_fused_moe_kernel(A: torch.Tensor,
827
compute_type=compute_type,
828
use_fp8_w8a8=use_fp8_w8a8,
829
use_int8_w8a16=use_int8_w8a16,
830
+ BLOCK_SIZE_K=BLOCK_SIZE_K,
831
**config,
832
)
833
vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 8,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 1,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 32,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 8,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 8,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 16,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 32,
123
+ "GROUP_SIZE_M": 1,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 32,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 16,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 8,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 1,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 32,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 8,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 8,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 16,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 32,
123
+ "GROUP_SIZE_M": 1,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 32,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 16,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 32,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 8,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 16,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 8,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 16,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 32,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 8,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 32,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 1,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 32,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 8,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 16,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 8,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 16,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 32,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 8,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 32,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 1,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 16,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 16,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 16,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 32,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 64,
123
+ "GROUP_SIZE_M": 8,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 64,
132
+ "GROUP_SIZE_M": 8,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 16,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 16,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 16,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 32,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 64,
123
+ "GROUP_SIZE_M": 8,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 64,
132
+ "GROUP_SIZE_M": 8,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 16,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 32,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 16,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 8,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 8,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 16,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 16,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 32,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 64,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 16,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 16,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 32,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 16,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 8,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 8,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 16,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 16,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 32,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 64,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 16,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 16,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 1,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 1,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 64,
69
+ "GROUP_SIZE_M": 32,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 64,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 128,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 128,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 128,
105
+ "GROUP_SIZE_M": 32,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 32,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 16,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 1,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 1,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 64,
69
+ "GROUP_SIZE_M": 32,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 64,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 128,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 128,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 128,
105
+ "GROUP_SIZE_M": 32,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 32,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 8,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 64,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 1,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 32,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 8,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 32,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 16,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 16,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 8,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 32,
123
+ "GROUP_SIZE_M": 8,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 32,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 32,
140
+ "BLOCK_SIZE_N": 32,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 32,
149
+ "BLOCK_SIZE_N": 32,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 32,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 8,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 64,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 1,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 32,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 8,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 32,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 16,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 16,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 8,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 32,
123
+ "GROUP_SIZE_M": 8,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 32,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 32,
140
+ "BLOCK_SIZE_N": 32,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 32,
149
+ "BLOCK_SIZE_N": 32,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 32,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 8,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 16,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 16,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 32,
123
+ "GROUP_SIZE_M": 1,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 8,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 16,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 16,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 32,
123
+ "GROUP_SIZE_M": 1,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 32,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 32,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 32,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 8,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 8,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 1,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 32,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 32,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 32,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 8,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 8,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 1,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 64,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 64,
15
+ "GROUP_SIZE_M": 1,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 64,
24
+ "GROUP_SIZE_M": 1,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 64,
33
+ "GROUP_SIZE_M": 1,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 64,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 64,
60
+ "GROUP_SIZE_M": 1,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 64,
69
+ "GROUP_SIZE_M": 1,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 64,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 64,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 64,
95
+ "BLOCK_SIZE_N": 128,
96
+ "GROUP_SIZE_M": 32,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 32,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 64,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 64,
15
+ "GROUP_SIZE_M": 1,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 64,
24
+ "GROUP_SIZE_M": 1,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 64,
33
+ "GROUP_SIZE_M": 1,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 64,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 64,
60
+ "GROUP_SIZE_M": 1,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 64,
69
+ "GROUP_SIZE_M": 1,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 64,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 64,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 64,
95
+ "BLOCK_SIZE_N": 128,
96
+ "GROUP_SIZE_M": 32,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 32,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=36864,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 128,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 128,
15
+ "GROUP_SIZE_M": 1,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 128,
24
+ "GROUP_SIZE_M": 1,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 8,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 128,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 128,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 64,
68
+ "BLOCK_SIZE_N": 128,
69
+ "GROUP_SIZE_M": 32,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 64,
77
+ "BLOCK_SIZE_N": 128,
78
+ "GROUP_SIZE_M": 32,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 128,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 64,
95
+ "BLOCK_SIZE_N": 128,
96
+ "GROUP_SIZE_M": 32,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 128,
105
+ "GROUP_SIZE_M": 32,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 32,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=36864,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 128,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 128,
15
+ "GROUP_SIZE_M": 1,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 128,
24
+ "GROUP_SIZE_M": 1,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 8,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 128,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 128,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 64,
68
+ "BLOCK_SIZE_N": 128,
69
+ "GROUP_SIZE_M": 32,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 64,
77
+ "BLOCK_SIZE_N": 128,
78
+ "GROUP_SIZE_M": 32,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 128,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 64,
95
+ "BLOCK_SIZE_N": 128,
96
+ "GROUP_SIZE_M": 32,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 128,
105
+ "GROUP_SIZE_M": 32,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 32,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 16,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 8,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 8,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 1,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 32,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 32,
132
+ "GROUP_SIZE_M": 8,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 16,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 8,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 8,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 1,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 32,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 32,
132
+ "GROUP_SIZE_M": 8,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 16,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 32,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 32,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 8,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 16,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 32,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 32,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 8,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 16,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 32,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 16,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 16,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 16,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 8,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 16,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 32,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 32,
132
+ "GROUP_SIZE_M": 16,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 32,
141
+ "GROUP_SIZE_M": 16,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 32,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 32,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 1,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 16,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 32,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 16,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 16,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 16,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 8,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 16,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 32,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 32,
132
+ "GROUP_SIZE_M": 16,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 32,
141
+ "GROUP_SIZE_M": 16,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 32,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 32,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 1,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 64,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 8,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 64,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 1,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 64,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 32,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 64,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 1,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 8,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 64,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 8,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 16,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 8,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 8,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 8,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 32,
123
+ "GROUP_SIZE_M": 1,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 32,
132
+ "GROUP_SIZE_M": 16,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 32,
141
+ "GROUP_SIZE_M": 16,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 32,
149
+ "BLOCK_SIZE_N": 64,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 32,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 64,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 8,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 64,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 1,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 64,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 32,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 64,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 1,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 8,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 64,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 8,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 16,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 8,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 8,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 8,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 32,
123
+ "GROUP_SIZE_M": 1,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 32,
132
+ "GROUP_SIZE_M": 16,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 32,
141
+ "GROUP_SIZE_M": 16,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 32,
149
+ "BLOCK_SIZE_N": 64,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 32,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 16,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 8,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 16,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 32,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 32,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 1,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 16,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 8,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 16,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 32,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 32,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 1,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 8,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 16,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 64,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 32,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 32,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 8,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 16,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 64,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 32,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 64,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 1,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 64,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 1,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 1,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 1,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 1,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 1,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 16,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 8,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 16,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 32,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 64,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 1,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 64,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 1,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 1,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 1,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 1,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 1,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 16,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 8,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 16,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 32,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 8,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 8,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 32,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 1,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 8,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 64,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 1,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 8,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 8,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 32,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 1,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 8,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 64,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 1,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 8,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 1,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 64,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 64,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 64,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 8,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 1,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 64,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 64,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 64,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 32,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 32,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 1,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 64,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 1,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 8,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 32,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 32,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 1,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 1,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 64,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 32,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 1,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 32,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 32,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 32,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 32,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 8,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 32,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 16,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 32,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 32,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 32,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 1,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 1,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 1,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 1,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 1,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 1,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 64,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 64,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 64,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 16,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 1,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 1,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 1,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 1,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 1,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 1,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 64,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 64,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 32,
104
+ "BLOCK_SIZE_N": 64,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 16,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=8192,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 16,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 16,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 16,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 32,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 8,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 32,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 64,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 64,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=7168,K=8192,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 16,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 16,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 16,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 32,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 8,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 16,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 32,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 8,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 64,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 64,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 32,
113
+ "BLOCK_SIZE_N": 128,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 32,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 1,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=8192,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 8,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 16,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 16,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 8,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 64,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file
vllm/model_executor/layers/quantization/utils/configs/N=8192,K=1536,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 128,
4
+ "BLOCK_SIZE_M": 32,
5
+ "BLOCK_SIZE_N": 32,
6
+ "GROUP_SIZE_M": 1,
7
+ "kpack": 1,
8
+ "matrix_instr_nonkdim": 16,
9
+ "num_warps": 4
10
+ },
11
+ "2": {
12
+ "BLOCK_SIZE_K": 128,
13
+ "BLOCK_SIZE_M": 32,
14
+ "BLOCK_SIZE_N": 32,
15
+ "GROUP_SIZE_M": 8,
16
+ "kpack": 1,
17
+ "matrix_instr_nonkdim": 16,
18
+ "num_warps": 4
19
+ },
20
+ "4": {
21
+ "BLOCK_SIZE_K": 128,
22
+ "BLOCK_SIZE_M": 32,
23
+ "BLOCK_SIZE_N": 32,
24
+ "GROUP_SIZE_M": 16,
25
+ "kpack": 1,
26
+ "matrix_instr_nonkdim": 16,
27
+ "num_warps": 4
28
+ },
29
+ "8": {
30
+ "BLOCK_SIZE_K": 128,
31
+ "BLOCK_SIZE_M": 32,
32
+ "BLOCK_SIZE_N": 32,
33
+ "GROUP_SIZE_M": 16,
34
+ "kpack": 1,
35
+ "matrix_instr_nonkdim": 16,
36
+ "num_warps": 4
37
+ },
38
+ "16": {
39
+ "BLOCK_SIZE_K": 128,
40
+ "BLOCK_SIZE_M": 32,
41
+ "BLOCK_SIZE_N": 32,
42
+ "GROUP_SIZE_M": 16,
43
+ "kpack": 1,
44
+ "matrix_instr_nonkdim": 16,
45
+ "num_warps": 4
46
+ },
47
+ "24": {
48
+ "BLOCK_SIZE_K": 128,
49
+ "BLOCK_SIZE_M": 32,
50
+ "BLOCK_SIZE_N": 32,
51
+ "GROUP_SIZE_M": 8,
52
+ "kpack": 1,
53
+ "matrix_instr_nonkdim": 16,
54
+ "num_warps": 4
55
+ },
56
+ "32": {
57
+ "BLOCK_SIZE_K": 128,
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "GROUP_SIZE_M": 8,
61
+ "kpack": 1,
62
+ "matrix_instr_nonkdim": 16,
63
+ "num_warps": 4
64
+ },
65
+ "48": {
66
+ "BLOCK_SIZE_K": 128,
67
+ "BLOCK_SIZE_M": 32,
68
+ "BLOCK_SIZE_N": 32,
69
+ "GROUP_SIZE_M": 8,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 128,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
78
+ "GROUP_SIZE_M": 16,
79
+ "kpack": 1,
80
+ "matrix_instr_nonkdim": 16,
81
+ "num_warps": 4
82
+ },
83
+ "96": {
84
+ "BLOCK_SIZE_K": 128,
85
+ "BLOCK_SIZE_M": 32,
86
+ "BLOCK_SIZE_N": 32,
87
+ "GROUP_SIZE_M": 1,
88
+ "kpack": 1,
89
+ "matrix_instr_nonkdim": 16,
90
+ "num_warps": 4
91
+ },
92
+ "128": {
93
+ "BLOCK_SIZE_K": 128,
94
+ "BLOCK_SIZE_M": 32,
95
+ "BLOCK_SIZE_N": 64,
96
+ "GROUP_SIZE_M": 1,
97
+ "kpack": 1,
98
+ "matrix_instr_nonkdim": 16,
99
+ "num_warps": 4
100
+ },
101
+ "256": {
102
+ "BLOCK_SIZE_K": 128,
103
+ "BLOCK_SIZE_M": 64,
104
+ "BLOCK_SIZE_N": 32,
105
+ "GROUP_SIZE_M": 1,
106
+ "kpack": 1,
107
+ "matrix_instr_nonkdim": 16,
108
+ "num_warps": 4
109
+ },
110
+ "512": {
111
+ "BLOCK_SIZE_K": 128,
112
+ "BLOCK_SIZE_M": 64,
113
+ "BLOCK_SIZE_N": 32,
114
+ "GROUP_SIZE_M": 1,
115
+ "kpack": 1,
116
+ "matrix_instr_nonkdim": 16,
117
+ "num_warps": 4
118
+ },
119
+ "1024": {
120
+ "BLOCK_SIZE_K": 128,
121
+ "BLOCK_SIZE_M": 64,
122
+ "BLOCK_SIZE_N": 128,
123
+ "GROUP_SIZE_M": 32,
124
+ "kpack": 1,
125
+ "matrix_instr_nonkdim": 16,
126
+ "num_warps": 4
127
+ },
128
+ "1536": {
129
+ "BLOCK_SIZE_K": 128,
130
+ "BLOCK_SIZE_M": 64,
131
+ "BLOCK_SIZE_N": 128,
132
+ "GROUP_SIZE_M": 32,
133
+ "kpack": 1,
134
+ "matrix_instr_nonkdim": 16,
135
+ "num_warps": 4
136
+ },
137
+ "2048": {
138
+ "BLOCK_SIZE_K": 128,
139
+ "BLOCK_SIZE_M": 64,
140
+ "BLOCK_SIZE_N": 128,
141
+ "GROUP_SIZE_M": 32,
142
+ "kpack": 1,
143
+ "matrix_instr_nonkdim": 16,
144
+ "num_warps": 4
145
+ },
146
+ "3072": {
147
+ "BLOCK_SIZE_K": 128,
148
+ "BLOCK_SIZE_M": 64,
149
+ "BLOCK_SIZE_N": 128,
150
+ "GROUP_SIZE_M": 32,
151
+ "kpack": 1,
152
+ "matrix_instr_nonkdim": 16,
153
+ "num_warps": 4
154
+ },
155
+ "4096": {
156
+ "BLOCK_SIZE_K": 128,
157
+ "BLOCK_SIZE_M": 64,
158
+ "BLOCK_SIZE_N": 128,
159
+ "GROUP_SIZE_M": 32,
160
+ "kpack": 1,
161
+ "matrix_instr_nonkdim": 16,
162
+ "num_warps": 4
163
+ }
164
+}
\ No newline at end of file