Hugging Face's logo

replicate
/
paged-attention

Kernels
Model card Files
xet
 Community
paged-attention
File size: 3,024 Bytes
132e594
6df83c0
ed30f9d
132e594
 
 
 
 
 
 
 
ed30f9d
132e594
4c6b316
132e594
 
 
6677800
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
132e594
 
ed30f9d
132e594
4c6b316
132e594
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4c6b316
132e594
 
6677800
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ed30f9d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
[general]
name = "paged_attention"
universal = false

[torch]
src = [
  "torch-ext/torch_binding.cpp",
  "torch-ext/torch_binding.h"
]

[kernel.cuda_utils]
backend = "cuda"
src = [
  "cuda-utils/cuda_utils.h",
  "cuda-utils/cuda_utils_kernels.cu",
]
depends = []

[kernel.cuda_utils_rocm]
backend = "rocm"
rocm-archs = [
    "gfx906",
    "gfx908",
    "gfx90a",
    "gfx940",
    "gfx941",
    "gfx942",
    "gfx1030",
    "gfx1100",
    "gfx1101",
]
src = [
  "cuda-utils/cuda_utils.h",
  "cuda-utils/cuda_utils_kernels.cu",
]
depends = ["torch"]

[kernel.paged_attention]
backend = "cuda"
src = [
  "cuda-utils/cuda_utils.h",
  "paged-attention/attention/attention_dtypes.h",
  "paged-attention/attention/attention_generic.cuh",
  "paged-attention/attention/attention_kernels.cuh",
  "paged-attention/attention/attention_utils.cuh",
  "paged-attention/attention/dtype_bfloat16.cuh",
  "paged-attention/attention/dtype_float16.cuh",
  "paged-attention/attention/dtype_float32.cuh",
  "paged-attention/attention/dtype_fp8.cuh",
  "paged-attention/attention/paged_attention_v1.cu",
  "paged-attention/attention/paged_attention_v2.cu",
  "paged-attention/cache_kernels.cu",
  "paged-attention/cuda_compat.h",
  "paged-attention/dispatch_utils.h",
  "paged-attention/quantization/fp8/amd/quant_utils.cuh",
  "paged-attention/quantization/fp8/nvidia/quant_utils.cuh",
]
include = [ "cuda-utils", "paged-attention" ]
depends = [ "torch" ]

[kernel.paged_attention_rocm]
backend = "rocm"
rocm-archs = [
    "gfx906",
    "gfx908",
    "gfx90a",
    "gfx940",
    "gfx941",
    "gfx942",
    "gfx1030",
    "gfx1100",
    "gfx1101",
]
src = [
  "cuda-utils/cuda_utils.h",
  "paged-attention/attention/attention_dtypes.h",
  "paged-attention/attention/attention_generic.cuh",
  "paged-attention/attention/attention_kernels.cuh",
  "paged-attention/attention/attention_utils.cuh",
  "paged-attention/attention/dtype_bfloat16.cuh",
  "paged-attention/attention/dtype_float16.cuh",
  "paged-attention/attention/dtype_float32.cuh",
  "paged-attention/attention/dtype_fp8.cuh",
  "paged-attention/attention/paged_attention_v1.cu",
  "paged-attention/attention/paged_attention_v2.cu",
  "paged-attention/cache_kernels.cu",
  "paged-attention/cuda_compat.h",
  "paged-attention/dispatch_utils.h",
  "paged-attention/quantization/fp8/amd/quant_utils.cuh",
  "paged-attention/quantization/fp8/nvidia/quant_utils.cuh",
]
include = [ "cuda-utils", "paged-attention" ]
depends = [ "torch" ]

[kernel.paged_attention_metal]
backend = "metal"
src = [
  "paged-attention-metal/attention/paged_attention.metal",
  "paged-attention-metal/cache/copy_blocks.metal",
  "paged-attention-metal/cache/reshape_and_cache.metal",
  "paged-attention-metal/convert_fp8.metal",
  "paged-attention-metal/float8.metal",
  "paged-attention-metal/utils.metal",
  "paged-attention-metal/paged_attention.mm",
  "paged-attention-metal/cache.mm",
  "paged-attention-metal/convert_fp8.mm",
  "paged-attention-metal/device.mm",
]
include = [ "." ]
depends = [ "torch" ]