sglang 0.4.2.post4__py3-none-any.whl → 0.4.3__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (57) hide show
  1. sglang/global_config.py +2 -0
  2. sglang/srt/entrypoints/engine.py +2 -2
  3. sglang/srt/layers/attention/flashinfer_backend.py +235 -110
  4. sglang/srt/layers/attention/triton_backend.py +358 -72
  5. sglang/srt/layers/attention/triton_ops/extend_attention.py +4 -4
  6. sglang/srt/layers/linear.py +12 -5
  7. sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json +2 -2
  8. sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
  9. sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +2 -2
  10. sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=14336,device_name=AMD_Instinct_MI325X.json +200 -0
  11. sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=1792,device_name=AMD_Instinct_MI325X.json +200 -0
  12. sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=3584,device_name=AMD_Instinct_MI325X.json +200 -0
  13. sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=4096,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +178 -0
  14. sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=7168,device_name=AMD_Instinct_MI325X.json +200 -0
  15. sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=8192,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +175 -0
  16. sglang/srt/layers/moe/fused_moe_triton/fused_moe.py +9 -2
  17. sglang/srt/layers/moe/fused_moe_triton/layer.py +2 -0
  18. sglang/srt/layers/moe/topk.py +1 -1
  19. sglang/srt/layers/quantization/__init__.py +51 -5
  20. sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
  21. sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +30 -30
  22. sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
  23. sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +29 -29
  24. sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
  25. sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +33 -33
  26. sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
  27. sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +31 -31
  28. sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
  29. sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +27 -27
  30. sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
  31. sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +31 -31
  32. sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
  33. sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +24 -24
  34. sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
  35. sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +30 -30
  36. sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
  37. sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +42 -42
  38. sglang/srt/layers/quantization/fp8_kernel.py +123 -17
  39. sglang/srt/layers/quantization/fp8_utils.py +33 -4
  40. sglang/srt/managers/detokenizer_manager.py +1 -0
  41. sglang/srt/managers/io_struct.py +4 -0
  42. sglang/srt/managers/schedule_batch.py +16 -3
  43. sglang/srt/managers/scheduler.py +29 -0
  44. sglang/srt/managers/tokenizer_manager.py +6 -0
  45. sglang/srt/managers/tp_worker_overlap_thread.py +4 -0
  46. sglang/srt/model_executor/cuda_graph_runner.py +12 -1
  47. sglang/srt/model_executor/model_runner.py +12 -2
  48. sglang/srt/models/deepseek_v2.py +17 -7
  49. sglang/srt/server_args.py +20 -1
  50. sglang/srt/speculative/eagle_worker.py +28 -8
  51. sglang/srt/utils.py +7 -0
  52. sglang/version.py +1 -1
  53. {sglang-0.4.2.post4.dist-info → sglang-0.4.3.dist-info}/METADATA +4 -3
  54. {sglang-0.4.2.post4.dist-info → sglang-0.4.3.dist-info}/RECORD +57 -41
  55. {sglang-0.4.2.post4.dist-info → sglang-0.4.3.dist-info}/LICENSE +0 -0
  56. {sglang-0.4.2.post4.dist-info → sglang-0.4.3.dist-info}/WHEEL +0 -0
  57. {sglang-0.4.2.post4.dist-info → sglang-0.4.3.dist-info}/top_level.txt +0 -0
@@ -0,0 +1,200 @@
1
+ {
2
+ "1": {
3
+ "BLOCK_SIZE_M": 16,
4
+ "BLOCK_SIZE_N": 16,
5
+ "BLOCK_SIZE_K": 128,
6
+ "GROUP_SIZE_M": 1,
7
+ "num_warps": 2,
8
+ "num_stages": 0,
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": 16,
16
+ "BLOCK_SIZE_K": 64,
17
+ "GROUP_SIZE_M": 1,
18
+ "num_warps": 2,
19
+ "num_stages": 0,
20
+ "waves_per_eu": 0,
21
+ "matrix_instr_nonkdim": 16,
22
+ "kpack": 2
23
+ },
24
+ "4": {
25
+ "BLOCK_SIZE_M": 16,
26
+ "BLOCK_SIZE_N": 32,
27
+ "BLOCK_SIZE_K": 256,
28
+ "GROUP_SIZE_M": 1,
29
+ "num_warps": 2,
30
+ "num_stages": 0,
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": 32,
38
+ "BLOCK_SIZE_K": 256,
39
+ "GROUP_SIZE_M": 1,
40
+ "num_warps": 2,
41
+ "num_stages": 0,
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": 32,
49
+ "BLOCK_SIZE_K": 256,
50
+ "GROUP_SIZE_M": 1,
51
+ "num_warps": 2,
52
+ "num_stages": 0,
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": 64,
61
+ "GROUP_SIZE_M": 1,
62
+ "num_warps": 4,
63
+ "num_stages": 0,
64
+ "waves_per_eu": 0,
65
+ "matrix_instr_nonkdim": 16,
66
+ "kpack": 1
67
+ },
68
+ "32": {
69
+ "BLOCK_SIZE_M": 16,
70
+ "BLOCK_SIZE_N": 16,
71
+ "BLOCK_SIZE_K": 256,
72
+ "GROUP_SIZE_M": 4,
73
+ "num_warps": 2,
74
+ "num_stages": 0,
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": 32,
82
+ "BLOCK_SIZE_K": 256,
83
+ "GROUP_SIZE_M": 1,
84
+ "num_warps": 2,
85
+ "num_stages": 0,
86
+ "waves_per_eu": 0,
87
+ "matrix_instr_nonkdim": 16,
88
+ "kpack": 2
89
+ },
90
+ "64": {
91
+ "BLOCK_SIZE_M": 32,
92
+ "BLOCK_SIZE_N": 32,
93
+ "BLOCK_SIZE_K": 256,
94
+ "GROUP_SIZE_M": 4,
95
+ "num_warps": 4,
96
+ "num_stages": 0,
97
+ "waves_per_eu": 0,
98
+ "matrix_instr_nonkdim": 16,
99
+ "kpack": 2
100
+ },
101
+ "96": {
102
+ "BLOCK_SIZE_M": 32,
103
+ "BLOCK_SIZE_N": 32,
104
+ "BLOCK_SIZE_K": 128,
105
+ "GROUP_SIZE_M": 4,
106
+ "num_warps": 4,
107
+ "num_stages": 0,
108
+ "waves_per_eu": 0,
109
+ "matrix_instr_nonkdim": 16,
110
+ "kpack": 1
111
+ },
112
+ "128": {
113
+ "BLOCK_SIZE_M": 64,
114
+ "BLOCK_SIZE_N": 64,
115
+ "BLOCK_SIZE_K": 128,
116
+ "GROUP_SIZE_M": 4,
117
+ "num_warps": 8,
118
+ "num_stages": 0,
119
+ "waves_per_eu": 0,
120
+ "matrix_instr_nonkdim": 16,
121
+ "kpack": 1
122
+ },
123
+ "256": {
124
+ "BLOCK_SIZE_M": 128,
125
+ "BLOCK_SIZE_N": 128,
126
+ "BLOCK_SIZE_K": 64,
127
+ "GROUP_SIZE_M": 4,
128
+ "num_warps": 8,
129
+ "num_stages": 0,
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": 128,
137
+ "BLOCK_SIZE_K": 64,
138
+ "GROUP_SIZE_M": 1,
139
+ "num_warps": 8,
140
+ "num_stages": 0,
141
+ "waves_per_eu": 0,
142
+ "matrix_instr_nonkdim": 32,
143
+ "kpack": 2
144
+ },
145
+ "1024": {
146
+ "BLOCK_SIZE_M": 128,
147
+ "BLOCK_SIZE_N": 128,
148
+ "BLOCK_SIZE_K": 64,
149
+ "GROUP_SIZE_M": 1,
150
+ "num_warps": 8,
151
+ "num_stages": 0,
152
+ "waves_per_eu": 0,
153
+ "matrix_instr_nonkdim": 16,
154
+ "kpack": 1
155
+ },
156
+ "1536": {
157
+ "BLOCK_SIZE_M": 128,
158
+ "BLOCK_SIZE_N": 128,
159
+ "BLOCK_SIZE_K": 64,
160
+ "GROUP_SIZE_M": 1,
161
+ "num_warps": 8,
162
+ "num_stages": 0,
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": 1,
172
+ "num_warps": 8,
173
+ "num_stages": 0,
174
+ "waves_per_eu": 0,
175
+ "matrix_instr_nonkdim": 16,
176
+ "kpack": 1
177
+ },
178
+ "3072": {
179
+ "BLOCK_SIZE_M": 128,
180
+ "BLOCK_SIZE_N": 128,
181
+ "BLOCK_SIZE_K": 64,
182
+ "GROUP_SIZE_M": 1,
183
+ "num_warps": 8,
184
+ "num_stages": 0,
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": 1,
194
+ "num_warps": 8,
195
+ "num_stages": 0,
196
+ "waves_per_eu": 0,
197
+ "matrix_instr_nonkdim": 16,
198
+ "kpack": 1
199
+ }
200
+ }
@@ -0,0 +1,178 @@
1
+ {
2
+ "4": {
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": 0,
9
+ "waves_per_eu": 4,
10
+ "matrix_instr_nonkdim": 16,
11
+ "kpack": 2
12
+ },
13
+ "8": {
14
+ "BLOCK_SIZE_M": 16,
15
+ "BLOCK_SIZE_N": 64,
16
+ "BLOCK_SIZE_K": 256,
17
+ "GROUP_SIZE_M": 1,
18
+ "num_warps": 4,
19
+ "num_stages": 0,
20
+ "waves_per_eu": 1,
21
+ "matrix_instr_nonkdim": 16,
22
+ "kpack": 1
23
+ },
24
+ "16": {
25
+ "BLOCK_SIZE_M": 32,
26
+ "BLOCK_SIZE_N": 64,
27
+ "BLOCK_SIZE_K": 256,
28
+ "GROUP_SIZE_M": 1,
29
+ "num_warps": 8,
30
+ "num_stages": 0,
31
+ "waves_per_eu": 2,
32
+ "matrix_instr_nonkdim": 16,
33
+ "kpack": 2
34
+ },
35
+ "32": {
36
+ "BLOCK_SIZE_M": 32,
37
+ "BLOCK_SIZE_N": 64,
38
+ "BLOCK_SIZE_K": 256,
39
+ "GROUP_SIZE_M": 1,
40
+ "num_warps": 8,
41
+ "num_stages": 0,
42
+ "waves_per_eu": 1,
43
+ "matrix_instr_nonkdim": 16,
44
+ "kpack": 2
45
+ },
46
+ "64": {
47
+ "BLOCK_SIZE_M": 32,
48
+ "BLOCK_SIZE_N": 64,
49
+ "BLOCK_SIZE_K": 256,
50
+ "GROUP_SIZE_M": 1,
51
+ "num_warps": 4,
52
+ "num_stages": 0,
53
+ "waves_per_eu": 2,
54
+ "matrix_instr_nonkdim": 16,
55
+ "kpack": 2
56
+ },
57
+ "128": {
58
+ "BLOCK_SIZE_M": 64,
59
+ "BLOCK_SIZE_N": 128,
60
+ "BLOCK_SIZE_K": 128,
61
+ "GROUP_SIZE_M": 1,
62
+ "num_warps": 8,
63
+ "num_stages": 0,
64
+ "waves_per_eu": 0,
65
+ "matrix_instr_nonkdim": 16,
66
+ "kpack": 1
67
+ },
68
+ "256": {
69
+ "BLOCK_SIZE_M": 64,
70
+ "BLOCK_SIZE_N": 128,
71
+ "BLOCK_SIZE_K": 128,
72
+ "GROUP_SIZE_M": 1,
73
+ "num_warps": 8,
74
+ "num_stages": 0,
75
+ "waves_per_eu": 0,
76
+ "matrix_instr_nonkdim": 16,
77
+ "kpack": 1
78
+ },
79
+ "512": {
80
+ "BLOCK_SIZE_M": 64,
81
+ "BLOCK_SIZE_N": 128,
82
+ "BLOCK_SIZE_K": 128,
83
+ "GROUP_SIZE_M": 1,
84
+ "num_warps": 8,
85
+ "num_stages": 0,
86
+ "waves_per_eu": 0,
87
+ "matrix_instr_nonkdim": 16,
88
+ "kpack": 2
89
+ },
90
+ "1024": {
91
+ "BLOCK_SIZE_M": 128,
92
+ "BLOCK_SIZE_N": 128,
93
+ "BLOCK_SIZE_K": 128,
94
+ "GROUP_SIZE_M": 1,
95
+ "num_warps": 8,
96
+ "num_stages": 0,
97
+ "waves_per_eu": 4,
98
+ "matrix_instr_nonkdim": 16,
99
+ "kpack": 2
100
+ },
101
+ "2048": {
102
+ "BLOCK_SIZE_M": 128,
103
+ "BLOCK_SIZE_N": 128,
104
+ "BLOCK_SIZE_K": 128,
105
+ "GROUP_SIZE_M": 1,
106
+ "num_warps": 8,
107
+ "num_stages": 0,
108
+ "waves_per_eu": 2,
109
+ "matrix_instr_nonkdim": 16,
110
+ "kpack": 2
111
+ },
112
+ "4096": {
113
+ "BLOCK_SIZE_M": 128,
114
+ "BLOCK_SIZE_N": 128,
115
+ "BLOCK_SIZE_K": 128,
116
+ "GROUP_SIZE_M": 1,
117
+ "num_warps": 8,
118
+ "num_stages": 0,
119
+ "waves_per_eu": 2,
120
+ "matrix_instr_nonkdim": 16,
121
+ "kpack": 2
122
+ },
123
+ "8192": {
124
+ "BLOCK_SIZE_M": 256,
125
+ "BLOCK_SIZE_N": 256,
126
+ "BLOCK_SIZE_K": 64,
127
+ "GROUP_SIZE_M": 1,
128
+ "num_warps": 8,
129
+ "num_stages": 0,
130
+ "waves_per_eu": 2,
131
+ "matrix_instr_nonkdim": 16,
132
+ "kpack": 1
133
+ },
134
+ "16384": {
135
+ "BLOCK_SIZE_M": 256,
136
+ "BLOCK_SIZE_N": 256,
137
+ "BLOCK_SIZE_K": 64,
138
+ "GROUP_SIZE_M": 1,
139
+ "num_warps": 8,
140
+ "num_stages": 0,
141
+ "waves_per_eu": 1,
142
+ "matrix_instr_nonkdim": 16,
143
+ "kpack": 1
144
+ },
145
+ "32768": {
146
+ "BLOCK_SIZE_M": 256,
147
+ "BLOCK_SIZE_N": 256,
148
+ "BLOCK_SIZE_K": 64,
149
+ "GROUP_SIZE_M": 1,
150
+ "num_warps": 8,
151
+ "num_stages": 0,
152
+ "waves_per_eu": 0,
153
+ "matrix_instr_nonkdim": 16,
154
+ "kpack": 1
155
+ },
156
+ "65536": {
157
+ "BLOCK_SIZE_M": 256,
158
+ "BLOCK_SIZE_N": 256,
159
+ "BLOCK_SIZE_K": 64,
160
+ "GROUP_SIZE_M": 1,
161
+ "num_warps": 8,
162
+ "num_stages": 0,
163
+ "waves_per_eu": 1,
164
+ "matrix_instr_nonkdim": 16,
165
+ "kpack": 1
166
+ },
167
+ "131072": {
168
+ "BLOCK_SIZE_M": 256,
169
+ "BLOCK_SIZE_N": 128,
170
+ "BLOCK_SIZE_K": 64,
171
+ "GROUP_SIZE_M": 1,
172
+ "num_warps": 4,
173
+ "num_stages": 0,
174
+ "waves_per_eu": 2,
175
+ "matrix_instr_nonkdim": 16,
176
+ "kpack": 2
177
+ }
178
+ }
@@ -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": 2,
8
+ "num_stages": 0,
9
+ "waves_per_eu": 0,
10
+ "matrix_instr_nonkdim": 16,
11
+ "kpack": 2
12
+ },
13
+ "2": {
14
+ "BLOCK_SIZE_M": 16,
15
+ "BLOCK_SIZE_N": 64,
16
+ "BLOCK_SIZE_K": 32,
17
+ "GROUP_SIZE_M": 1,
18
+ "num_warps": 4,
19
+ "num_stages": 0,
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": 32,
27
+ "BLOCK_SIZE_K": 128,
28
+ "GROUP_SIZE_M": 1,
29
+ "num_warps": 4,
30
+ "num_stages": 0,
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": 32,
38
+ "BLOCK_SIZE_K": 256,
39
+ "GROUP_SIZE_M": 1,
40
+ "num_warps": 2,
41
+ "num_stages": 0,
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": 16,
49
+ "BLOCK_SIZE_K": 256,
50
+ "GROUP_SIZE_M": 1,
51
+ "num_warps": 4,
52
+ "num_stages": 0,
53
+ "waves_per_eu": 0,
54
+ "matrix_instr_nonkdim": 16,
55
+ "kpack": 2
56
+ },
57
+ "24": {
58
+ "BLOCK_SIZE_M": 32,
59
+ "BLOCK_SIZE_N": 32,
60
+ "BLOCK_SIZE_K": 128,
61
+ "GROUP_SIZE_M": 1,
62
+ "num_warps": 8,
63
+ "num_stages": 0,
64
+ "waves_per_eu": 0,
65
+ "matrix_instr_nonkdim": 16,
66
+ "kpack": 1
67
+ },
68
+ "32": {
69
+ "BLOCK_SIZE_M": 16,
70
+ "BLOCK_SIZE_N": 32,
71
+ "BLOCK_SIZE_K": 128,
72
+ "GROUP_SIZE_M": 4,
73
+ "num_warps": 2,
74
+ "num_stages": 0,
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": 32,
82
+ "BLOCK_SIZE_K": 128,
83
+ "GROUP_SIZE_M": 4,
84
+ "num_warps": 2,
85
+ "num_stages": 0,
86
+ "waves_per_eu": 0,
87
+ "matrix_instr_nonkdim": 16,
88
+ "kpack": 1
89
+ },
90
+ "64": {
91
+ "BLOCK_SIZE_M": 32,
92
+ "BLOCK_SIZE_N": 32,
93
+ "BLOCK_SIZE_K": 128,
94
+ "GROUP_SIZE_M": 4,
95
+ "num_warps": 4,
96
+ "num_stages": 0,
97
+ "waves_per_eu": 0,
98
+ "matrix_instr_nonkdim": 16,
99
+ "kpack": 2
100
+ },
101
+ "96": {
102
+ "BLOCK_SIZE_M": 32,
103
+ "BLOCK_SIZE_N": 32,
104
+ "BLOCK_SIZE_K": 128,
105
+ "GROUP_SIZE_M": 4,
106
+ "num_warps": 4,
107
+ "num_stages": 0,
108
+ "waves_per_eu": 0,
109
+ "matrix_instr_nonkdim": 16,
110
+ "kpack": 2
111
+ },
112
+ "128": {
113
+ "BLOCK_SIZE_M": 64,
114
+ "BLOCK_SIZE_N": 64,
115
+ "BLOCK_SIZE_K": 64,
116
+ "GROUP_SIZE_M": 4,
117
+ "num_warps": 8,
118
+ "num_stages": 0,
119
+ "waves_per_eu": 0,
120
+ "matrix_instr_nonkdim": 16,
121
+ "kpack": 1
122
+ },
123
+ "256": {
124
+ "BLOCK_SIZE_M": 128,
125
+ "BLOCK_SIZE_N": 128,
126
+ "BLOCK_SIZE_K": 64,
127
+ "GROUP_SIZE_M": 4,
128
+ "num_warps": 8,
129
+ "num_stages": 0,
130
+ "waves_per_eu": 0,
131
+ "matrix_instr_nonkdim": 32,
132
+ "kpack": 2
133
+ },
134
+ "512": {
135
+ "BLOCK_SIZE_M": 128,
136
+ "BLOCK_SIZE_N": 128,
137
+ "BLOCK_SIZE_K": 64,
138
+ "GROUP_SIZE_M": 1,
139
+ "num_warps": 8,
140
+ "num_stages": 0,
141
+ "waves_per_eu": 0,
142
+ "matrix_instr_nonkdim": 16,
143
+ "kpack": 1
144
+ },
145
+ "1024": {
146
+ "BLOCK_SIZE_M": 128,
147
+ "BLOCK_SIZE_N": 128,
148
+ "BLOCK_SIZE_K": 64,
149
+ "GROUP_SIZE_M": 1,
150
+ "num_warps": 8,
151
+ "num_stages": 0,
152
+ "waves_per_eu": 0,
153
+ "matrix_instr_nonkdim": 16,
154
+ "kpack": 1
155
+ },
156
+ "1536": {
157
+ "BLOCK_SIZE_M": 128,
158
+ "BLOCK_SIZE_N": 128,
159
+ "BLOCK_SIZE_K": 64,
160
+ "GROUP_SIZE_M": 1,
161
+ "num_warps": 8,
162
+ "num_stages": 0,
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": 1,
172
+ "num_warps": 8,
173
+ "num_stages": 0,
174
+ "waves_per_eu": 0,
175
+ "matrix_instr_nonkdim": 16,
176
+ "kpack": 1
177
+ },
178
+ "3072": {
179
+ "BLOCK_SIZE_M": 128,
180
+ "BLOCK_SIZE_N": 128,
181
+ "BLOCK_SIZE_K": 64,
182
+ "GROUP_SIZE_M": 1,
183
+ "num_warps": 8,
184
+ "num_stages": 0,
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": 1,
194
+ "num_warps": 8,
195
+ "num_stages": 0,
196
+ "waves_per_eu": 0,
197
+ "matrix_instr_nonkdim": 16,
198
+ "kpack": 1
199
+ }
200
+ }