sglang 0.4.2.post3__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.
- sglang/check_env.py +1 -0
- sglang/global_config.py +2 -0
- sglang/srt/constrained/outlines_backend.py +4 -1
- sglang/srt/entrypoints/engine.py +2 -2
- sglang/srt/layers/attention/flashinfer_backend.py +265 -147
- sglang/srt/layers/attention/triton_backend.py +358 -72
- sglang/srt/layers/attention/triton_ops/extend_attention.py +4 -4
- sglang/srt/layers/linear.py +12 -5
- 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
- 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
- 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
- sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=14336,device_name=AMD_Instinct_MI325X.json +200 -0
- sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=1792,device_name=AMD_Instinct_MI325X.json +200 -0
- sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=3584,device_name=AMD_Instinct_MI325X.json +200 -0
- sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=4096,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +178 -0
- sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=7168,device_name=AMD_Instinct_MI325X.json +200 -0
- sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=8192,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json +175 -0
- sglang/srt/layers/moe/fused_moe_triton/fused_moe.py +27 -5
- sglang/srt/layers/moe/fused_moe_triton/layer.py +2 -0
- sglang/srt/layers/moe/topk.py +1 -1
- sglang/srt/layers/quantization/__init__.py +51 -5
- sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
- sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
- sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +30 -30
- sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
- sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
- sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
- sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
- sglang/srt/layers/quantization/configs/N=256,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
- sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
- sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +29 -29
- sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
- sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
- sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +33 -33
- sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
- sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +31 -31
- sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
- sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +27 -27
- sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
- sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +31 -31
- sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
- sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
- sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
- sglang/srt/layers/quantization/configs/N=7168,K=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
- sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
- sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
- sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
- sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +24 -24
- sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
- sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +30 -30
- sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json +164 -0
- sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json +42 -42
- sglang/srt/layers/quantization/fp8_kernel.py +123 -17
- sglang/srt/layers/quantization/fp8_utils.py +33 -4
- sglang/srt/lora/backend/__init__.py +25 -5
- sglang/srt/lora/backend/base_backend.py +31 -9
- sglang/srt/lora/backend/flashinfer_backend.py +41 -4
- sglang/srt/lora/backend/triton_backend.py +34 -4
- sglang/srt/lora/layers.py +293 -0
- sglang/srt/lora/lora.py +101 -326
- sglang/srt/lora/lora_manager.py +101 -269
- sglang/srt/lora/mem_pool.py +174 -0
- sglang/srt/lora/triton_ops/__init__.py +7 -1
- sglang/srt/lora/triton_ops/gate_up_lora_b.py +170 -0
- sglang/srt/lora/triton_ops/qkv_lora_b.py +5 -5
- sglang/srt/lora/triton_ops/sgemm_lora_a.py +2 -2
- sglang/srt/lora/triton_ops/sgemm_lora_b.py +2 -2
- sglang/srt/lora/utils.py +141 -0
- sglang/srt/managers/detokenizer_manager.py +1 -0
- sglang/srt/managers/io_struct.py +4 -0
- sglang/srt/managers/schedule_batch.py +16 -3
- sglang/srt/managers/scheduler.py +29 -0
- sglang/srt/managers/tokenizer_manager.py +6 -0
- sglang/srt/managers/tp_worker_overlap_thread.py +4 -0
- sglang/srt/model_executor/cuda_graph_runner.py +16 -1
- sglang/srt/model_executor/model_runner.py +12 -2
- sglang/srt/models/deepseek_v2.py +17 -7
- sglang/srt/server_args.py +20 -1
- sglang/srt/speculative/eagle_draft_cuda_graph_runner.py +1 -0
- sglang/srt/speculative/eagle_utils.py +64 -21
- sglang/srt/speculative/eagle_worker.py +29 -8
- sglang/srt/utils.py +7 -0
- sglang/version.py +1 -1
- {sglang-0.4.2.post3.dist-info → sglang-0.4.3.dist-info}/METADATA +6 -5
- {sglang-0.4.2.post3.dist-info → sglang-0.4.3.dist-info}/RECORD +88 -55
- {sglang-0.4.2.post3.dist-info → sglang-0.4.3.dist-info}/LICENSE +0 -0
- {sglang-0.4.2.post3.dist-info → sglang-0.4.3.dist-info}/WHEEL +0 -0
- {sglang-0.4.2.post3.dist-info → sglang-0.4.3.dist-info}/top_level.txt +0 -0
@@ -0,0 +1,164 @@
|
|
1
|
+
{
|
2
|
+
"1": {
|
3
|
+
"BLOCK_SIZE_M": 32,
|
4
|
+
"BLOCK_SIZE_N": 32,
|
5
|
+
"BLOCK_SIZE_K": 128,
|
6
|
+
"GROUP_SIZE_M": 16,
|
7
|
+
"num_warps": 4,
|
8
|
+
"num_stages": 2,
|
9
|
+
"waves_per_eu": 0
|
10
|
+
},
|
11
|
+
"2": {
|
12
|
+
"BLOCK_SIZE_M": 32,
|
13
|
+
"BLOCK_SIZE_N": 64,
|
14
|
+
"BLOCK_SIZE_K": 128,
|
15
|
+
"GROUP_SIZE_M": 1,
|
16
|
+
"num_warps": 4,
|
17
|
+
"num_stages": 2,
|
18
|
+
"waves_per_eu": 0
|
19
|
+
},
|
20
|
+
"4": {
|
21
|
+
"BLOCK_SIZE_M": 64,
|
22
|
+
"BLOCK_SIZE_N": 64,
|
23
|
+
"BLOCK_SIZE_K": 128,
|
24
|
+
"GROUP_SIZE_M": 16,
|
25
|
+
"num_warps": 4,
|
26
|
+
"num_stages": 2,
|
27
|
+
"waves_per_eu": 0
|
28
|
+
},
|
29
|
+
"8": {
|
30
|
+
"BLOCK_SIZE_M": 32,
|
31
|
+
"BLOCK_SIZE_N": 128,
|
32
|
+
"BLOCK_SIZE_K": 128,
|
33
|
+
"GROUP_SIZE_M": 32,
|
34
|
+
"num_warps": 4,
|
35
|
+
"num_stages": 2,
|
36
|
+
"waves_per_eu": 0
|
37
|
+
},
|
38
|
+
"16": {
|
39
|
+
"BLOCK_SIZE_M": 32,
|
40
|
+
"BLOCK_SIZE_N": 128,
|
41
|
+
"BLOCK_SIZE_K": 128,
|
42
|
+
"GROUP_SIZE_M": 1,
|
43
|
+
"num_warps": 4,
|
44
|
+
"num_stages": 2,
|
45
|
+
"waves_per_eu": 0
|
46
|
+
},
|
47
|
+
"24": {
|
48
|
+
"BLOCK_SIZE_M": 32,
|
49
|
+
"BLOCK_SIZE_N": 128,
|
50
|
+
"BLOCK_SIZE_K": 128,
|
51
|
+
"GROUP_SIZE_M": 4,
|
52
|
+
"num_warps": 4,
|
53
|
+
"num_stages": 2,
|
54
|
+
"waves_per_eu": 0
|
55
|
+
},
|
56
|
+
"32": {
|
57
|
+
"BLOCK_SIZE_M": 32,
|
58
|
+
"BLOCK_SIZE_N": 128,
|
59
|
+
"BLOCK_SIZE_K": 128,
|
60
|
+
"GROUP_SIZE_M": 8,
|
61
|
+
"num_warps": 4,
|
62
|
+
"num_stages": 2,
|
63
|
+
"waves_per_eu": 0
|
64
|
+
},
|
65
|
+
"48": {
|
66
|
+
"BLOCK_SIZE_M": 32,
|
67
|
+
"BLOCK_SIZE_N": 128,
|
68
|
+
"BLOCK_SIZE_K": 128,
|
69
|
+
"GROUP_SIZE_M": 4,
|
70
|
+
"num_warps": 4,
|
71
|
+
"num_stages": 2,
|
72
|
+
"waves_per_eu": 0
|
73
|
+
},
|
74
|
+
"64": {
|
75
|
+
"BLOCK_SIZE_M": 32,
|
76
|
+
"BLOCK_SIZE_N": 128,
|
77
|
+
"BLOCK_SIZE_K": 128,
|
78
|
+
"GROUP_SIZE_M": 4,
|
79
|
+
"num_warps": 4,
|
80
|
+
"num_stages": 2,
|
81
|
+
"waves_per_eu": 0
|
82
|
+
},
|
83
|
+
"96": {
|
84
|
+
"BLOCK_SIZE_M": 32,
|
85
|
+
"BLOCK_SIZE_N": 128,
|
86
|
+
"BLOCK_SIZE_K": 128,
|
87
|
+
"GROUP_SIZE_M": 8,
|
88
|
+
"num_warps": 4,
|
89
|
+
"num_stages": 2,
|
90
|
+
"waves_per_eu": 0
|
91
|
+
},
|
92
|
+
"128": {
|
93
|
+
"BLOCK_SIZE_M": 32,
|
94
|
+
"BLOCK_SIZE_N": 16,
|
95
|
+
"BLOCK_SIZE_K": 128,
|
96
|
+
"GROUP_SIZE_M": 4,
|
97
|
+
"num_warps": 4,
|
98
|
+
"num_stages": 2,
|
99
|
+
"waves_per_eu": 0
|
100
|
+
},
|
101
|
+
"256": {
|
102
|
+
"BLOCK_SIZE_M": 64,
|
103
|
+
"BLOCK_SIZE_N": 16,
|
104
|
+
"BLOCK_SIZE_K": 128,
|
105
|
+
"GROUP_SIZE_M": 1,
|
106
|
+
"num_warps": 4,
|
107
|
+
"num_stages": 2,
|
108
|
+
"waves_per_eu": 0
|
109
|
+
},
|
110
|
+
"512": {
|
111
|
+
"BLOCK_SIZE_M": 64,
|
112
|
+
"BLOCK_SIZE_N": 64,
|
113
|
+
"BLOCK_SIZE_K": 128,
|
114
|
+
"GROUP_SIZE_M": 32,
|
115
|
+
"num_warps": 4,
|
116
|
+
"num_stages": 2,
|
117
|
+
"waves_per_eu": 0
|
118
|
+
},
|
119
|
+
"1024": {
|
120
|
+
"BLOCK_SIZE_M": 64,
|
121
|
+
"BLOCK_SIZE_N": 64,
|
122
|
+
"BLOCK_SIZE_K": 128,
|
123
|
+
"GROUP_SIZE_M": 4,
|
124
|
+
"num_warps": 8,
|
125
|
+
"num_stages": 2,
|
126
|
+
"waves_per_eu": 0
|
127
|
+
},
|
128
|
+
"1536": {
|
129
|
+
"BLOCK_SIZE_M": 64,
|
130
|
+
"BLOCK_SIZE_N": 64,
|
131
|
+
"BLOCK_SIZE_K": 128,
|
132
|
+
"GROUP_SIZE_M": 8,
|
133
|
+
"num_warps": 4,
|
134
|
+
"num_stages": 2,
|
135
|
+
"waves_per_eu": 0
|
136
|
+
},
|
137
|
+
"2048": {
|
138
|
+
"BLOCK_SIZE_M": 32,
|
139
|
+
"BLOCK_SIZE_N": 64,
|
140
|
+
"BLOCK_SIZE_K": 128,
|
141
|
+
"GROUP_SIZE_M": 1,
|
142
|
+
"num_warps": 4,
|
143
|
+
"num_stages": 2,
|
144
|
+
"waves_per_eu": 0
|
145
|
+
},
|
146
|
+
"3072": {
|
147
|
+
"BLOCK_SIZE_M": 32,
|
148
|
+
"BLOCK_SIZE_N": 128,
|
149
|
+
"BLOCK_SIZE_K": 128,
|
150
|
+
"GROUP_SIZE_M": 1,
|
151
|
+
"num_warps": 4,
|
152
|
+
"num_stages": 2,
|
153
|
+
"waves_per_eu": 0
|
154
|
+
},
|
155
|
+
"4096": {
|
156
|
+
"BLOCK_SIZE_M": 64,
|
157
|
+
"BLOCK_SIZE_N": 128,
|
158
|
+
"BLOCK_SIZE_K": 64,
|
159
|
+
"GROUP_SIZE_M": 4,
|
160
|
+
"num_warps": 4,
|
161
|
+
"num_stages": 2,
|
162
|
+
"waves_per_eu": 0
|
163
|
+
}
|
164
|
+
}
|
sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=14336,device_name=AMD_Instinct_MI325X.json
ADDED
@@ -0,0 +1,200 @@
|
|
1
|
+
{
|
2
|
+
"1": {
|
3
|
+
"BLOCK_SIZE_M": 16,
|
4
|
+
"BLOCK_SIZE_N": 32,
|
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": 1
|
12
|
+
},
|
13
|
+
"2": {
|
14
|
+
"BLOCK_SIZE_M": 16,
|
15
|
+
"BLOCK_SIZE_N": 16,
|
16
|
+
"BLOCK_SIZE_K": 128,
|
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": 16,
|
38
|
+
"BLOCK_SIZE_K": 256,
|
39
|
+
"GROUP_SIZE_M": 1,
|
40
|
+
"num_warps": 1,
|
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": 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": 16,
|
59
|
+
"BLOCK_SIZE_N": 32,
|
60
|
+
"BLOCK_SIZE_K": 64,
|
61
|
+
"GROUP_SIZE_M": 1,
|
62
|
+
"num_warps": 1,
|
63
|
+
"num_stages": 0,
|
64
|
+
"waves_per_eu": 0,
|
65
|
+
"matrix_instr_nonkdim": 16,
|
66
|
+
"kpack": 2
|
67
|
+
},
|
68
|
+
"32": {
|
69
|
+
"BLOCK_SIZE_M": 16,
|
70
|
+
"BLOCK_SIZE_N": 16,
|
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": 1
|
78
|
+
},
|
79
|
+
"48": {
|
80
|
+
"BLOCK_SIZE_M": 16,
|
81
|
+
"BLOCK_SIZE_N": 16,
|
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": 2
|
89
|
+
},
|
90
|
+
"64": {
|
91
|
+
"BLOCK_SIZE_M": 32,
|
92
|
+
"BLOCK_SIZE_N": 64,
|
93
|
+
"BLOCK_SIZE_K": 128,
|
94
|
+
"GROUP_SIZE_M": 4,
|
95
|
+
"num_warps": 8,
|
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": 2
|
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": 128,
|
136
|
+
"BLOCK_SIZE_N": 128,
|
137
|
+
"BLOCK_SIZE_K": 64,
|
138
|
+
"GROUP_SIZE_M": 4,
|
139
|
+
"num_warps": 8,
|
140
|
+
"num_stages": 0,
|
141
|
+
"waves_per_eu": 0,
|
142
|
+
"matrix_instr_nonkdim": 16,
|
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": 32,
|
154
|
+
"kpack": 2
|
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": 2
|
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": 1
|
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
|
+
}
|
sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=1792,device_name=AMD_Instinct_MI325X.json
ADDED
@@ -0,0 +1,200 @@
|
|
1
|
+
{
|
2
|
+
"1": {
|
3
|
+
"BLOCK_SIZE_M": 16,
|
4
|
+
"BLOCK_SIZE_N": 32,
|
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": 128,
|
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": 64,
|
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": 2
|
34
|
+
},
|
35
|
+
"8": {
|
36
|
+
"BLOCK_SIZE_M": 16,
|
37
|
+
"BLOCK_SIZE_N": 16,
|
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": 64,
|
48
|
+
"BLOCK_SIZE_N": 64,
|
49
|
+
"BLOCK_SIZE_K": 64,
|
50
|
+
"GROUP_SIZE_M": 1,
|
51
|
+
"num_warps": 8,
|
52
|
+
"num_stages": 0,
|
53
|
+
"waves_per_eu": 0,
|
54
|
+
"matrix_instr_nonkdim": 16,
|
55
|
+
"kpack": 1
|
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": 128,
|
82
|
+
"BLOCK_SIZE_K": 64,
|
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
|
+
"64": {
|
91
|
+
"BLOCK_SIZE_M": 16,
|
92
|
+
"BLOCK_SIZE_N": 64,
|
93
|
+
"BLOCK_SIZE_K": 64,
|
94
|
+
"GROUP_SIZE_M": 1,
|
95
|
+
"num_warps": 2,
|
96
|
+
"num_stages": 0,
|
97
|
+
"waves_per_eu": 0,
|
98
|
+
"matrix_instr_nonkdim": 16,
|
99
|
+
"kpack": 1
|
100
|
+
},
|
101
|
+
"96": {
|
102
|
+
"BLOCK_SIZE_M": 32,
|
103
|
+
"BLOCK_SIZE_N": 32,
|
104
|
+
"BLOCK_SIZE_K": 256,
|
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": 16,
|
132
|
+
"kpack": 1
|
133
|
+
},
|
134
|
+
"512": {
|
135
|
+
"BLOCK_SIZE_M": 64,
|
136
|
+
"BLOCK_SIZE_N": 64,
|
137
|
+
"BLOCK_SIZE_K": 128,
|
138
|
+
"GROUP_SIZE_M": 1,
|
139
|
+
"num_warps": 8,
|
140
|
+
"num_stages": 0,
|
141
|
+
"waves_per_eu": 0,
|
142
|
+
"matrix_instr_nonkdim": 16,
|
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": 1
|
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": 2
|
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": 1
|
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
|
+
}
|