Small build fixes
Browse files- build.toml +45 -43
- flake.lock +3 -3
- flash-attn/flash_api.cpp +1 -1
build.toml
CHANGED
|
@@ -1,6 +1,8 @@
|
|
| 1 |
[general]
|
| 2 |
-
name = "
|
| 3 |
universal = false
|
|
|
|
|
|
|
| 4 |
|
| 5 |
[torch]
|
| 6 |
src = [
|
|
@@ -25,7 +27,7 @@ cuda-flags = [
|
|
| 25 |
"--use_fast_math",
|
| 26 |
"-DNDEBUG",
|
| 27 |
]
|
| 28 |
-
|
| 29 |
src = [
|
| 30 |
"flash-attn/cuda_check.h",
|
| 31 |
"flash-attn/flash_api.cpp",
|
|
@@ -57,7 +59,6 @@ cuda-flags = [
|
|
| 57 |
"--expt-extended-lambda",
|
| 58 |
"--use_fast_math",
|
| 59 |
"-DNDEBUG",
|
| 60 |
-
"-DFLASHATTENTION_DISABLE_PYBIND",
|
| 61 |
]
|
| 62 |
src = [
|
| 63 |
"flash-attn/block.h",
|
|
@@ -87,6 +88,7 @@ src = [
|
|
| 87 |
"flash-attn/softmax.h",
|
| 88 |
"flash-attn/tile_size.h",
|
| 89 |
"flash-attn/tile_scheduler.hpp",
|
|
|
|
| 90 |
"flash-attn/instantiations/flash_bwd_hdim128_bf16_sm80.cu",
|
| 91 |
"flash-attn/instantiations/flash_bwd_hdim128_bf16_softcap_sm80.cu",
|
| 92 |
"flash-attn/instantiations/flash_bwd_hdim128_fp16_sm80.cu",
|
|
@@ -108,85 +110,85 @@ src = [
|
|
| 108 |
"flash-attn/instantiations/flash_bwd_hdim96_fp16_sm80.cu",
|
| 109 |
"flash-attn/instantiations/flash_bwd_hdim96_fp16_softcap_sm80.cu",
|
| 110 |
"flash-attn/instantiations/flash_fwd_hdim128_bf16_paged_sm80.cu",
|
| 111 |
-
"flash-attn/instantiations/
|
| 112 |
"flash-attn/instantiations/flash_fwd_hdim128_bf16_paged_split_sm80.cu",
|
| 113 |
-
"flash-attn/instantiations/
|
| 114 |
"flash-attn/instantiations/flash_fwd_hdim128_bf16_sm80.cu",
|
| 115 |
-
"flash-attn/instantiations/
|
| 116 |
"flash-attn/instantiations/flash_fwd_hdim128_bf16_split_sm80.cu",
|
| 117 |
-
"flash-attn/instantiations/
|
| 118 |
"flash-attn/instantiations/flash_fwd_hdim128_fp16_paged_sm80.cu",
|
| 119 |
-
"flash-attn/instantiations/
|
| 120 |
"flash-attn/instantiations/flash_fwd_hdim128_fp16_paged_split_sm80.cu",
|
| 121 |
-
"flash-attn/instantiations/
|
| 122 |
"flash-attn/instantiations/flash_fwd_hdim128_fp16_sm80.cu",
|
| 123 |
-
"flash-attn/instantiations/
|
| 124 |
"flash-attn/instantiations/flash_fwd_hdim128_fp16_split_sm80.cu",
|
| 125 |
-
"flash-attn/instantiations/
|
| 126 |
"flash-attn/instantiations/flash_fwd_hdim192_bf16_paged_sm80.cu",
|
| 127 |
-
"flash-attn/instantiations/
|
| 128 |
"flash-attn/instantiations/flash_fwd_hdim192_bf16_paged_split_sm80.cu",
|
| 129 |
-
"flash-attn/instantiations/
|
| 130 |
"flash-attn/instantiations/flash_fwd_hdim192_bf16_sm80.cu",
|
| 131 |
-
"flash-attn/instantiations/
|
| 132 |
"flash-attn/instantiations/flash_fwd_hdim192_bf16_split_sm80.cu",
|
| 133 |
-
"flash-attn/instantiations/
|
| 134 |
"flash-attn/instantiations/flash_fwd_hdim192_fp16_paged_sm80.cu",
|
| 135 |
-
"flash-attn/instantiations/
|
| 136 |
"flash-attn/instantiations/flash_fwd_hdim192_fp16_paged_split_sm80.cu",
|
| 137 |
-
"flash-attn/instantiations/
|
| 138 |
"flash-attn/instantiations/flash_fwd_hdim192_fp16_sm80.cu",
|
| 139 |
-
"flash-attn/instantiations/
|
| 140 |
"flash-attn/instantiations/flash_fwd_hdim192_fp16_split_sm80.cu",
|
| 141 |
-
"flash-attn/instantiations/
|
| 142 |
"flash-attn/instantiations/flash_fwd_hdim256_bf16_paged_sm80.cu",
|
| 143 |
-
"flash-attn/instantiations/
|
| 144 |
"flash-attn/instantiations/flash_fwd_hdim256_bf16_paged_split_sm80.cu",
|
| 145 |
-
"flash-attn/instantiations/
|
| 146 |
"flash-attn/instantiations/flash_fwd_hdim256_bf16_sm80.cu",
|
| 147 |
-
"flash-attn/instantiations/
|
| 148 |
"flash-attn/instantiations/flash_fwd_hdim256_bf16_split_sm80.cu",
|
| 149 |
-
"flash-attn/instantiations/
|
| 150 |
"flash-attn/instantiations/flash_fwd_hdim256_fp16_paged_sm80.cu",
|
| 151 |
-
"flash-attn/instantiations/
|
| 152 |
"flash-attn/instantiations/flash_fwd_hdim256_fp16_paged_split_sm80.cu",
|
| 153 |
-
"flash-attn/instantiations/
|
| 154 |
"flash-attn/instantiations/flash_fwd_hdim256_fp16_sm80.cu",
|
| 155 |
-
"flash-attn/instantiations/
|
| 156 |
"flash-attn/instantiations/flash_fwd_hdim256_fp16_split_sm80.cu",
|
| 157 |
-
"flash-attn/instantiations/
|
| 158 |
"flash-attn/instantiations/flash_fwd_hdim64_bf16_paged_sm80.cu",
|
| 159 |
-
"flash-attn/instantiations/
|
| 160 |
"flash-attn/instantiations/flash_fwd_hdim64_bf16_paged_split_sm80.cu",
|
| 161 |
-
"flash-attn/instantiations/
|
| 162 |
"flash-attn/instantiations/flash_fwd_hdim64_bf16_sm80.cu",
|
| 163 |
-
"flash-attn/instantiations/
|
| 164 |
"flash-attn/instantiations/flash_fwd_hdim64_bf16_split_sm80.cu",
|
| 165 |
-
"flash-attn/instantiations/
|
| 166 |
"flash-attn/instantiations/flash_fwd_hdim64_fp16_paged_sm80.cu",
|
| 167 |
-
"flash-attn/instantiations/
|
| 168 |
"flash-attn/instantiations/flash_fwd_hdim64_fp16_paged_split_sm80.cu",
|
| 169 |
-
"flash-attn/instantiations/
|
| 170 |
"flash-attn/instantiations/flash_fwd_hdim64_fp16_sm80.cu",
|
| 171 |
-
"flash-attn/instantiations/
|
| 172 |
"flash-attn/instantiations/flash_fwd_hdim64_fp16_split_sm80.cu",
|
| 173 |
-
"flash-attn/instantiations/
|
| 174 |
"flash-attn/instantiations/flash_fwd_hdim96_bf16_paged_sm80.cu",
|
| 175 |
-
"flash-attn/instantiations/
|
| 176 |
"flash-attn/instantiations/flash_fwd_hdim96_bf16_paged_split_sm80.cu",
|
| 177 |
-
"flash-attn/instantiations/
|
| 178 |
"flash-attn/instantiations/flash_fwd_hdim96_bf16_sm80.cu",
|
| 179 |
-
"flash-attn/instantiations/
|
| 180 |
"flash-attn/instantiations/flash_fwd_hdim96_bf16_split_sm80.cu",
|
| 181 |
-
"flash-attn/instantiations/
|
| 182 |
"flash-attn/instantiations/flash_fwd_hdim96_fp16_paged_sm80.cu",
|
| 183 |
-
"flash-attn/instantiations/
|
| 184 |
"flash-attn/instantiations/flash_fwd_hdim96_fp16_paged_split_sm80.cu",
|
| 185 |
-
"flash-attn/instantiations/
|
| 186 |
"flash-attn/instantiations/flash_fwd_hdim96_fp16_sm80.cu",
|
| 187 |
-
"flash-attn/instantiations/
|
| 188 |
"flash-attn/instantiations/flash_fwd_hdim96_fp16_split_sm80.cu",
|
| 189 |
-
"flash-attn/instantiations/
|
| 190 |
]
|
| 191 |
include = ["flash-attn"]
|
| 192 |
depends = ["torch", "cutlass_3_9"]
|
|
|
|
| 1 |
[general]
|
| 2 |
+
name = "vllm_flash_attn3"
|
| 3 |
universal = false
|
| 4 |
+
cuda-minver = "12.4"
|
| 5 |
+
cuda-maxver = "12.4"
|
| 6 |
|
| 7 |
[torch]
|
| 8 |
src = [
|
|
|
|
| 27 |
"--use_fast_math",
|
| 28 |
"-DNDEBUG",
|
| 29 |
]
|
| 30 |
+
cxx-flags = ["-DFLASHATTENTION_DISABLE_PYBIND"]
|
| 31 |
src = [
|
| 32 |
"flash-attn/cuda_check.h",
|
| 33 |
"flash-attn/flash_api.cpp",
|
|
|
|
| 59 |
"--expt-extended-lambda",
|
| 60 |
"--use_fast_math",
|
| 61 |
"-DNDEBUG",
|
|
|
|
| 62 |
]
|
| 63 |
src = [
|
| 64 |
"flash-attn/block.h",
|
|
|
|
| 88 |
"flash-attn/softmax.h",
|
| 89 |
"flash-attn/tile_size.h",
|
| 90 |
"flash-attn/tile_scheduler.hpp",
|
| 91 |
+
|
| 92 |
"flash-attn/instantiations/flash_bwd_hdim128_bf16_sm80.cu",
|
| 93 |
"flash-attn/instantiations/flash_bwd_hdim128_bf16_softcap_sm80.cu",
|
| 94 |
"flash-attn/instantiations/flash_bwd_hdim128_fp16_sm80.cu",
|
|
|
|
| 110 |
"flash-attn/instantiations/flash_bwd_hdim96_fp16_sm80.cu",
|
| 111 |
"flash-attn/instantiations/flash_bwd_hdim96_fp16_softcap_sm80.cu",
|
| 112 |
"flash-attn/instantiations/flash_fwd_hdim128_bf16_paged_sm80.cu",
|
| 113 |
+
"flash-attn/instantiations/flash_fwd_hdim128_bf16_paged_softcap_sm80.cu",
|
| 114 |
"flash-attn/instantiations/flash_fwd_hdim128_bf16_paged_split_sm80.cu",
|
| 115 |
+
"flash-attn/instantiations/flash_fwd_hdim128_bf16_paged_split_softcap_sm80.cu",
|
| 116 |
"flash-attn/instantiations/flash_fwd_hdim128_bf16_sm80.cu",
|
| 117 |
+
"flash-attn/instantiations/flash_fwd_hdim128_bf16_softcap_sm80.cu",
|
| 118 |
"flash-attn/instantiations/flash_fwd_hdim128_bf16_split_sm80.cu",
|
| 119 |
+
"flash-attn/instantiations/flash_fwd_hdim128_bf16_split_softcap_sm80.cu",
|
| 120 |
"flash-attn/instantiations/flash_fwd_hdim128_fp16_paged_sm80.cu",
|
| 121 |
+
"flash-attn/instantiations/flash_fwd_hdim128_fp16_paged_softcap_sm80.cu",
|
| 122 |
"flash-attn/instantiations/flash_fwd_hdim128_fp16_paged_split_sm80.cu",
|
| 123 |
+
"flash-attn/instantiations/flash_fwd_hdim128_fp16_paged_split_softcap_sm80.cu",
|
| 124 |
"flash-attn/instantiations/flash_fwd_hdim128_fp16_sm80.cu",
|
| 125 |
+
"flash-attn/instantiations/flash_fwd_hdim128_fp16_softcap_sm80.cu",
|
| 126 |
"flash-attn/instantiations/flash_fwd_hdim128_fp16_split_sm80.cu",
|
| 127 |
+
"flash-attn/instantiations/flash_fwd_hdim128_fp16_split_softcap_sm80.cu",
|
| 128 |
"flash-attn/instantiations/flash_fwd_hdim192_bf16_paged_sm80.cu",
|
| 129 |
+
"flash-attn/instantiations/flash_fwd_hdim192_bf16_paged_softcap_sm80.cu",
|
| 130 |
"flash-attn/instantiations/flash_fwd_hdim192_bf16_paged_split_sm80.cu",
|
| 131 |
+
"flash-attn/instantiations/flash_fwd_hdim192_bf16_paged_split_softcap_sm80.cu",
|
| 132 |
"flash-attn/instantiations/flash_fwd_hdim192_bf16_sm80.cu",
|
| 133 |
+
"flash-attn/instantiations/flash_fwd_hdim192_bf16_softcap_sm80.cu",
|
| 134 |
"flash-attn/instantiations/flash_fwd_hdim192_bf16_split_sm80.cu",
|
| 135 |
+
"flash-attn/instantiations/flash_fwd_hdim192_bf16_split_softcap_sm80.cu",
|
| 136 |
"flash-attn/instantiations/flash_fwd_hdim192_fp16_paged_sm80.cu",
|
| 137 |
+
"flash-attn/instantiations/flash_fwd_hdim192_fp16_paged_softcap_sm80.cu",
|
| 138 |
"flash-attn/instantiations/flash_fwd_hdim192_fp16_paged_split_sm80.cu",
|
| 139 |
+
"flash-attn/instantiations/flash_fwd_hdim192_fp16_paged_split_softcap_sm80.cu",
|
| 140 |
"flash-attn/instantiations/flash_fwd_hdim192_fp16_sm80.cu",
|
| 141 |
+
"flash-attn/instantiations/flash_fwd_hdim192_fp16_softcap_sm80.cu",
|
| 142 |
"flash-attn/instantiations/flash_fwd_hdim192_fp16_split_sm80.cu",
|
| 143 |
+
"flash-attn/instantiations/flash_fwd_hdim192_fp16_split_softcap_sm80.cu",
|
| 144 |
"flash-attn/instantiations/flash_fwd_hdim256_bf16_paged_sm80.cu",
|
| 145 |
+
"flash-attn/instantiations/flash_fwd_hdim256_bf16_paged_softcap_sm80.cu",
|
| 146 |
"flash-attn/instantiations/flash_fwd_hdim256_bf16_paged_split_sm80.cu",
|
| 147 |
+
"flash-attn/instantiations/flash_fwd_hdim256_bf16_paged_split_softcap_sm80.cu",
|
| 148 |
"flash-attn/instantiations/flash_fwd_hdim256_bf16_sm80.cu",
|
| 149 |
+
"flash-attn/instantiations/flash_fwd_hdim256_bf16_softcap_sm80.cu",
|
| 150 |
"flash-attn/instantiations/flash_fwd_hdim256_bf16_split_sm80.cu",
|
| 151 |
+
"flash-attn/instantiations/flash_fwd_hdim256_bf16_split_softcap_sm80.cu",
|
| 152 |
"flash-attn/instantiations/flash_fwd_hdim256_fp16_paged_sm80.cu",
|
| 153 |
+
"flash-attn/instantiations/flash_fwd_hdim256_fp16_paged_softcap_sm80.cu",
|
| 154 |
"flash-attn/instantiations/flash_fwd_hdim256_fp16_paged_split_sm80.cu",
|
| 155 |
+
"flash-attn/instantiations/flash_fwd_hdim256_fp16_paged_split_softcap_sm80.cu",
|
| 156 |
"flash-attn/instantiations/flash_fwd_hdim256_fp16_sm80.cu",
|
| 157 |
+
"flash-attn/instantiations/flash_fwd_hdim256_fp16_softcap_sm80.cu",
|
| 158 |
"flash-attn/instantiations/flash_fwd_hdim256_fp16_split_sm80.cu",
|
| 159 |
+
"flash-attn/instantiations/flash_fwd_hdim256_fp16_split_softcap_sm80.cu",
|
| 160 |
"flash-attn/instantiations/flash_fwd_hdim64_bf16_paged_sm80.cu",
|
| 161 |
+
"flash-attn/instantiations/flash_fwd_hdim64_bf16_paged_softcap_sm80.cu",
|
| 162 |
"flash-attn/instantiations/flash_fwd_hdim64_bf16_paged_split_sm80.cu",
|
| 163 |
+
"flash-attn/instantiations/flash_fwd_hdim64_bf16_paged_split_softcap_sm80.cu",
|
| 164 |
"flash-attn/instantiations/flash_fwd_hdim64_bf16_sm80.cu",
|
| 165 |
+
"flash-attn/instantiations/flash_fwd_hdim64_bf16_softcap_sm80.cu",
|
| 166 |
"flash-attn/instantiations/flash_fwd_hdim64_bf16_split_sm80.cu",
|
| 167 |
+
"flash-attn/instantiations/flash_fwd_hdim64_bf16_split_softcap_sm80.cu",
|
| 168 |
"flash-attn/instantiations/flash_fwd_hdim64_fp16_paged_sm80.cu",
|
| 169 |
+
"flash-attn/instantiations/flash_fwd_hdim64_fp16_paged_softcap_sm80.cu",
|
| 170 |
"flash-attn/instantiations/flash_fwd_hdim64_fp16_paged_split_sm80.cu",
|
| 171 |
+
"flash-attn/instantiations/flash_fwd_hdim64_fp16_paged_split_softcap_sm80.cu",
|
| 172 |
"flash-attn/instantiations/flash_fwd_hdim64_fp16_sm80.cu",
|
| 173 |
+
"flash-attn/instantiations/flash_fwd_hdim64_fp16_softcap_sm80.cu",
|
| 174 |
"flash-attn/instantiations/flash_fwd_hdim64_fp16_split_sm80.cu",
|
| 175 |
+
"flash-attn/instantiations/flash_fwd_hdim64_fp16_split_softcap_sm80.cu",
|
| 176 |
"flash-attn/instantiations/flash_fwd_hdim96_bf16_paged_sm80.cu",
|
| 177 |
+
"flash-attn/instantiations/flash_fwd_hdim96_bf16_paged_softcap_sm80.cu",
|
| 178 |
"flash-attn/instantiations/flash_fwd_hdim96_bf16_paged_split_sm80.cu",
|
| 179 |
+
"flash-attn/instantiations/flash_fwd_hdim96_bf16_paged_split_softcap_sm80.cu",
|
| 180 |
"flash-attn/instantiations/flash_fwd_hdim96_bf16_sm80.cu",
|
| 181 |
+
"flash-attn/instantiations/flash_fwd_hdim96_bf16_softcap_sm80.cu",
|
| 182 |
"flash-attn/instantiations/flash_fwd_hdim96_bf16_split_sm80.cu",
|
| 183 |
+
"flash-attn/instantiations/flash_fwd_hdim96_bf16_split_softcap_sm80.cu",
|
| 184 |
"flash-attn/instantiations/flash_fwd_hdim96_fp16_paged_sm80.cu",
|
| 185 |
+
"flash-attn/instantiations/flash_fwd_hdim96_fp16_paged_softcap_sm80.cu",
|
| 186 |
"flash-attn/instantiations/flash_fwd_hdim96_fp16_paged_split_sm80.cu",
|
| 187 |
+
"flash-attn/instantiations/flash_fwd_hdim96_fp16_paged_split_softcap_sm80.cu",
|
| 188 |
"flash-attn/instantiations/flash_fwd_hdim96_fp16_sm80.cu",
|
| 189 |
+
"flash-attn/instantiations/flash_fwd_hdim96_fp16_softcap_sm80.cu",
|
| 190 |
"flash-attn/instantiations/flash_fwd_hdim96_fp16_split_sm80.cu",
|
| 191 |
+
"flash-attn/instantiations/flash_fwd_hdim96_fp16_split_softcap_sm80.cu"
|
| 192 |
]
|
| 193 |
include = ["flash-attn"]
|
| 194 |
depends = ["torch", "cutlass_3_9"]
|
flake.lock
CHANGED
|
@@ -98,11 +98,11 @@
|
|
| 98 |
]
|
| 99 |
},
|
| 100 |
"locked": {
|
| 101 |
-
"lastModified":
|
| 102 |
-
"narHash": "sha256-
|
| 103 |
"owner": "huggingface",
|
| 104 |
"repo": "kernel-builder",
|
| 105 |
-
"rev": "
|
| 106 |
"type": "github"
|
| 107 |
},
|
| 108 |
"original": {
|
|
|
|
| 98 |
]
|
| 99 |
},
|
| 100 |
"locked": {
|
| 101 |
+
"lastModified": 1750860635,
|
| 102 |
+
"narHash": "sha256-Wzz7UQRuzSAQCvmPclYN/xmAqpMR19QDgmsCzknAzeo=",
|
| 103 |
"owner": "huggingface",
|
| 104 |
"repo": "kernel-builder",
|
| 105 |
+
"rev": "7e2d7303d9f2d49d470bddcbb8379fffeee38dcf",
|
| 106 |
"type": "github"
|
| 107 |
},
|
| 108 |
"original": {
|
flash-attn/flash_api.cpp
CHANGED
|
@@ -1620,4 +1620,4 @@ PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
|
|
| 1620 |
m.def("get_scheduler_metadata", &mha_fwd_get_scheduler_metadata, "Get scheduler metadata for varlen forward pass");
|
| 1621 |
}
|
| 1622 |
|
| 1623 |
-
#endif
|
|
|
|
| 1620 |
m.def("get_scheduler_metadata", &mha_fwd_get_scheduler_metadata, "Get scheduler metadata for varlen forward pass");
|
| 1621 |
}
|
| 1622 |
|
| 1623 |
+
#endif
|