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, 23:30 UTC
[AMD] [Model] DeepSeek tunings (#13199)
rasmith committed
Feb 15, 2025 at 05:58 UTC
ed0de3e4b8439513a7dc211ef459910c9b91ff92
24 files changed
+3936
vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=AMD_Instinct_MI300X,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": 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": 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": 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": 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": 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": 64,
131
+ "BLOCK_SIZE_N": 64,
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": 64,
141
+ "GROUP_SIZE_M": 8,
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": 64,
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_MI300X,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": 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": 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": 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": 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": 32,
113
+ "BLOCK_SIZE_N": 64,
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": 64,
122
+ "BLOCK_SIZE_N": 64,
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": 64,
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": 64,
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_MI300X,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": 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": 64,
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": 64,
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": 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": 64,
68
+ "BLOCK_SIZE_N": 32,
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": 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": 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": 64,
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": 32,
104
+ "BLOCK_SIZE_N": 64,
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": 128,
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": 64,
131
+ "BLOCK_SIZE_N": 64,
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=2304,K=7168,device_name=AMD_Instinct_MI300X,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": 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": 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": 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": 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": 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": 64,
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": 64,
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": 64,
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": 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": 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_MI300X,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": 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": 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": 64,
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": 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": 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": 64,
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": 64,
122
+ "BLOCK_SIZE_N": 64,
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": 64,
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": 64,
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": 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": 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_MI300X,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": 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": 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": 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": 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": 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": 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": 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": 32,
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": 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_MI300X,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": 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": 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": 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": 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": 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": 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": 64,
122
+ "BLOCK_SIZE_N": 64,
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": 64,
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_MI300X,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": 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": 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": 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": 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": 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": 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": 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": 64,
122
+ "BLOCK_SIZE_N": 64,
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": 64,
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": 64,
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": 64,
150
+ "GROUP_SIZE_M": 1,
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_MI300X,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": 64,
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": 64,
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": 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": 64,
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": 64,
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": 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": 64,
68
+ "BLOCK_SIZE_N": 64,
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": 64,
77
+ "BLOCK_SIZE_N": 64,
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": 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": 64,
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": 64,
122
+ "BLOCK_SIZE_N": 64,
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": 64,
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=36864,K=7168,device_name=AMD_Instinct_MI300X,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": 64,
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": 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": 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": 64,
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": 128,
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": 128,
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": 64,
68
+ "BLOCK_SIZE_N": 64,
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": 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": 64,
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": 64,
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": 64,
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": 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": 64,
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": 64,
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_MI300X,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": 64,
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": 64,
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": 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": 64,
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": 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": 32,
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": 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": 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": 128,
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": 128,
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": 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": 64,
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=4608,K=7168,device_name=AMD_Instinct_MI300X,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": 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": 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": 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": 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": 64,
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": 64,
122
+ "BLOCK_SIZE_N": 64,
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": 64,
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": 64,
150
+ "GROUP_SIZE_M": 1,
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_MI300X,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": 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": 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": 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": 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": 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": 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": 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": 64,
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": 64,
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_MI300X,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": 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": 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": 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": 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": 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": 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": 64,
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": 64,
159
+ "GROUP_SIZE_M": 8,
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_MI300X,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": 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": 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": 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": 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": 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": 64,
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": 64,
122
+ "BLOCK_SIZE_N": 64,
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": 64,
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": 64,
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": 64,
150
+ "GROUP_SIZE_M": 1,
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": 64,
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=7168,K=1152,device_name=AMD_Instinct_MI300X,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": 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": 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": 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": 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": 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": 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": 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": 64,
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": 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": 64,
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": 64,
150
+ "GROUP_SIZE_M": 1,
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": 64,
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=7168,K=128,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 256,
4
+ "BLOCK_SIZE_M": 64,
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": 64,
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": 64,
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": 64,
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": 64,
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": 128,
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": 64,
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": 128,
69
+ "GROUP_SIZE_M": 16,
70
+ "kpack": 1,
71
+ "matrix_instr_nonkdim": 16,
72
+ "num_warps": 4
73
+ },
74
+ "64": {
75
+ "BLOCK_SIZE_K": 256,
76
+ "BLOCK_SIZE_M": 32,
77
+ "BLOCK_SIZE_N": 32,
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": 64,
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": 128,
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": 64,
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": 128,
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": 64,
122
+ "BLOCK_SIZE_N": 64,
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": 64,
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=7168,K=16384,device_name=AMD_Instinct_MI300X,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": 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": 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": 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": 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": 64,
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": 64,
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": 64,
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": 64,
122
+ "BLOCK_SIZE_N": 64,
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": 64,
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_MI300X,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": 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": 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": 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": 64,
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": 64,
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": 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": 64,
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_MI300X,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": 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": 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": 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": 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": 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": 64,
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": 64,
122
+ "BLOCK_SIZE_N": 64,
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": 64,
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": 64,
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": 64,
150
+ "GROUP_SIZE_M": 1,
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_MI300X,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": 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": 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": 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": 32,
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": 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": 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": 64,
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": 64,
122
+ "BLOCK_SIZE_N": 64,
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": 64,
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": 64,
150
+ "GROUP_SIZE_M": 1,
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_MI300X,dtype=fp8_w8a8,block_shape=[128,128].json
new
+164
@@ -0,0 +1,164 @@
1
+{
2
+ "1": {
3
+ "BLOCK_SIZE_K": 32,
4
+ "BLOCK_SIZE_M": 256,
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": 64,
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": 64,
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": 64,
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": 64,
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": 64,
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": 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": 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": 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": 128,
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": 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": 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": 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": 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": 64,
159
+ "GROUP_SIZE_M": 16,
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_MI300X,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": 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": 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": 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": 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": 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": 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": 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": 64,
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": 64,
122
+ "BLOCK_SIZE_N": 64,
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": 64,
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": 64,
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": 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": 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_MI300X,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": 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": 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": 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": 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": 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": 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": 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": 64,
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": 64,
122
+ "BLOCK_SIZE_N": 64,
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": 64,
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": 64,
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": 64,
150
+ "GROUP_SIZE_M": 1,
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": 64,
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