forked from tile-ai/TileOPs
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy path__init__.py
More file actions
202 lines (200 loc) · 5.29 KB
/
Copy path__init__.py
File metadata and controls
202 lines (200 loc) · 5.29 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
from .attention import (
DeepSeekSparseAttentionDecodeWithKVCacheFwdOp,
GroupedQueryAttentionBwdOp,
GroupedQueryAttentionDecodePagedWithKVCacheFwdOp,
GroupedQueryAttentionDecodeWithKVCacheFwdOp,
GroupedQueryAttentionFwdOp,
GroupedQueryAttentionPrefillFwdOp,
GroupedQueryAttentionPrefillPagedWithKVCacheFwdOp,
GroupedQueryAttentionPrefillVarlenFwdOp,
GroupedQueryAttentionPrefillWithKVCacheFwdOp,
GroupedQueryAttentionSlidingWindowFwdOp,
GroupedQueryAttentionSlidingWindowVarlenFwdOp,
MeanPoolingForwardOp,
MultiHeadAttentionBwdOp,
MultiHeadAttentionDecodePagedWithKVCacheFwdOp,
MultiHeadAttentionDecodeWithKVCacheFwdOp,
MultiHeadAttentionFwdOp,
MultiHeadLatentAttentionDecodeWithKVCacheFwdOp,
NSACmpFwdVarlenOp,
NSAFwdVarlenOp,
NSATopkVarlenOp,
)
from .convolution import Conv1dBiasFwdOp, Conv1dFwdOp, Conv2dOp, Conv3dOp
from .da_cumsum import DaCumsumFwdOp
from .deltanet import DeltaNetBwdOp, DeltaNetFwdOp, DeltaNetOp
from .deltanet_recurrence import DeltaNetDecodeOp
from .dropout import DropoutOp
from .elementwise import BinaryOp, FusedGatedOp, UnaryOp
from .fft import FFTC2COp
from .fp8_lighting_indexer import FP8LightingIndexerOp
from .fp8_quant import FP8QuantOp
from .gated_deltanet import (
GatedDeltaNetBwdOp,
GatedDeltaNetDecodeOp,
GatedDeltaNetFwdOp,
GatedDeltaNetOp,
)
from .gated_linear_attn import GLADecodeOp
from .gemm import GemmOp
from .gla import GLABwdOp, GLAFwdOp
from .grouped_gemm import GroupedGemmOp
from .mhc import MHCPostOp, MHCPreOp
from .moe import MoePermuteAlignFwdOp
from .norm import (
AdaLayerNormFwdOp,
AdaLayerNormZeroFwdOp,
BatchNormBwdOp,
BatchNormFwdOp,
FusedAddLayerNormFwdOp,
FusedAddRMSNormFwdOp,
GroupNormFwdOp,
InstanceNormFwdOp,
LayerNormFwdOp,
RMSNormFwdOp,
)
from .op_base import Op
from .pool import AvgPool1dOp, AvgPool2dOp, AvgPool3dOp
# --- Reduction ops (uncomment as sub-category PRs land) ---
from .reduction import (
AllFwdOp,
AmaxFwdOp, # ReduceMaxOp
AminFwdOp, # ReduceMinOp
AnyFwdOp,
ArgmaxFwdOp,
ArgminFwdOp,
CountNonzeroFwdOp,
# CummaxOp,
# CumminOp,
CumprodFwdOp,
CumsumFwdOp,
InfNormFwdOp,
L1NormFwdOp,
L2NormFwdOp,
LogSoftmaxFwdOp,
LogSumExpFwdOp,
MeanFwdOp, # ReduceMeanOp
ProdFwdOp, # ReduceProdOp
SoftmaxFwdOp,
StdFwdOp,
SumFwdOp, # ReduceSumOp
VarFwdOp,
VarMeanFwdOp,
)
from .rope import (
RopeLlama31Op,
RopeLongRopeOp,
RopeNeoxOp,
RopeNeoxPositionIdsOp,
RopeNonNeoxOp,
RopeYarnOp,
)
from .ssd_chunk_scan import SSDChunkScanFwdOp
from .ssd_chunk_state import SSDChunkStateFwdOp
from .ssd_decode import SSDDecodeOp
from .ssd_state_passing import SSDStatePassingFwdOp
from .topk_selector import TopkSelectorOp
__all__ = [
"BinaryOp",
"AvgPool1dOp",
"AvgPool2dOp",
"AvgPool3dOp",
"AdaLayerNormFwdOp",
"AdaLayerNormZeroFwdOp",
"BatchNormBwdOp",
"BatchNormFwdOp",
"Conv1dBiasFwdOp",
"Conv1dFwdOp",
"Conv2dOp",
"Conv3dOp",
"DaCumsumFwdOp",
"DeepSeekSparseAttentionDecodeWithKVCacheFwdOp",
"DropoutOp",
"FFTC2COp",
"FP8LightingIndexerOp",
"FP8QuantOp",
"FusedAddLayerNormFwdOp",
"FusedAddRMSNormFwdOp",
"FusedGatedOp",
"DeltaNetBwdOp",
"DeltaNetDecodeOp",
"DeltaNetFwdOp",
"DeltaNetOp",
"GatedDeltaNetBwdOp",
"GatedDeltaNetDecodeOp",
"GatedDeltaNetFwdOp",
"GatedDeltaNetOp",
"GLABwdOp",
"GLADecodeOp",
"GLAFwdOp",
"GemmOp",
"GroupedQueryAttentionSlidingWindowFwdOp",
"GroupedQueryAttentionSlidingWindowVarlenFwdOp",
"GroupedQueryAttentionBwdOp",
"GroupedQueryAttentionDecodePagedWithKVCacheFwdOp",
"GroupedQueryAttentionDecodeWithKVCacheFwdOp",
"GroupedQueryAttentionFwdOp",
"GroupedQueryAttentionPrefillFwdOp",
"GroupedQueryAttentionPrefillPagedWithKVCacheFwdOp",
"GroupedQueryAttentionPrefillVarlenFwdOp",
"GroupedQueryAttentionPrefillWithKVCacheFwdOp",
"GroupNormFwdOp",
"GroupedGemmOp",
"InstanceNormFwdOp",
"LayerNormFwdOp",
"MHCPostOp",
"MHCPreOp",
"MeanPoolingForwardOp",
"MultiHeadAttentionBwdOp",
"MultiHeadAttentionDecodePagedWithKVCacheFwdOp",
"MultiHeadAttentionDecodeWithKVCacheFwdOp",
"MultiHeadAttentionFwdOp",
"MultiHeadLatentAttentionDecodeWithKVCacheFwdOp",
"NSACmpFwdVarlenOp",
"NSAFwdVarlenOp",
"NSATopkVarlenOp",
"Op",
"MoePermuteAlignFwdOp",
"RMSNormFwdOp",
"SSDChunkScanFwdOp",
"SSDChunkStateFwdOp",
"SSDDecodeOp",
"SSDStatePassingFwdOp",
"RopeLlama31Op",
"RopeLongRopeOp",
"RopeNeoxOp",
"RopeNeoxPositionIdsOp",
"RopeNonNeoxOp",
"RopeYarnOp",
"UnaryOp",
"TopkSelectorOp",
# --- Reduction ops (uncomment as sub-category PRs land) ---
"AllFwdOp",
"AmaxFwdOp",
"AminFwdOp",
"AnyFwdOp",
"ArgmaxFwdOp",
"ArgminFwdOp",
"CountNonzeroFwdOp",
# "CummaxOp",
# "CumminOp",
"CumprodFwdOp",
"CumsumFwdOp",
"InfNormFwdOp",
"L1NormFwdOp",
"L2NormFwdOp",
"LogSoftmaxFwdOp",
"LogSumExpFwdOp",
"MeanFwdOp",
"ProdFwdOp",
# "ReduceMaxOp",
# "ReduceMeanOp",
# "ReduceMinOp",
# "ReduceProdOp",
# "ReduceSumOp",
"SoftmaxFwdOp",
"StdFwdOp",
"SumFwdOp",
"VarMeanFwdOp",
"VarFwdOp",
]