danieldk HF staff commited on
Commit
4fcc03d
·
verified ·
1 Parent(s): a923ada

Add activation kernels

Browse files
README.md CHANGED
@@ -1,3 +1,3 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
1
+ ## Activation
2
+
3
+ Activation kernels from [vLLM](https://github.com/vllm-project/vllm/blob/main/csrc/activation_kernels.cu).
activation/activation_kernels.cu ADDED
@@ -0,0 +1,204 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #include <ATen/cuda/CUDAContext.h>
2
+ #include <torch/all.h>
3
+ #include <c10/cuda/CUDAGuard.h>
4
+
5
+ #include <cmath>
6
+
7
+ #include "cuda_compat.h"
8
+ #include "dispatch_utils.h"
9
+
10
+ namespace vllm {
11
+
12
+ // Activation and gating kernel template.
13
+ template <typename scalar_t, scalar_t (*ACT_FN)(const scalar_t&)>
14
+ __global__ void act_and_mul_kernel(
15
+ scalar_t* __restrict__ out, // [..., d]
16
+ const scalar_t* __restrict__ input, // [..., 2, d]
17
+ const int d) {
18
+ const int64_t token_idx = blockIdx.x;
19
+ for (int64_t idx = threadIdx.x; idx < d; idx += blockDim.x) {
20
+ const scalar_t x = VLLM_LDG(&input[token_idx * 2 * d + idx]);
21
+ const scalar_t y = VLLM_LDG(&input[token_idx * 2 * d + d + idx]);
22
+ out[token_idx * d + idx] = ACT_FN(x) * y;
23
+ }
24
+ }
25
+
26
+ template <typename T>
27
+ __device__ __forceinline__ T silu_kernel(const T& x) {
28
+ // x * sigmoid(x)
29
+ return (T)(((float)x) / (1.0f + expf((float)-x)));
30
+ }
31
+
32
+ template <typename T>
33
+ __device__ __forceinline__ T gelu_kernel(const T& x) {
34
+ // Equivalent to PyTorch GELU with 'none' approximation.
35
+ // Refer to:
36
+ // https://github.com/pytorch/pytorch/blob/8ac9b20d4b090c213799e81acf48a55ea8d437d6/aten/src/ATen/native/cuda/ActivationGeluKernel.cu#L36-L38
37
+ const float f = (float)x;
38
+ constexpr float ALPHA = M_SQRT1_2;
39
+ return (T)(f * 0.5f * (1.0f + ::erf(f * ALPHA)));
40
+ }
41
+
42
+ template <typename T>
43
+ __device__ __forceinline__ T gelu_tanh_kernel(const T& x) {
44
+ // Equivalent to PyTorch GELU with 'tanh' approximation.
45
+ // Refer to:
46
+ // https://github.com/pytorch/pytorch/blob/8ac9b20d4b090c213799e81acf48a55ea8d437d6/aten/src/ATen/native/cuda/ActivationGeluKernel.cu#L25-L30
47
+ const float f = (float)x;
48
+ constexpr float BETA = M_SQRT2 * M_2_SQRTPI * 0.5f;
49
+ constexpr float KAPPA = 0.044715;
50
+ float x_cube = f * f * f;
51
+ float inner = BETA * (f + KAPPA * x_cube);
52
+ return (T)(0.5f * f * (1.0f + ::tanhf(inner)));
53
+ }
54
+
55
+ } // namespace vllm
56
+
57
+ // Launch activation and gating kernel.
58
+ #define LAUNCH_ACTIVATION_GATE_KERNEL(KERNEL) \
59
+ int d = input.size(-1) / 2; \
60
+ int64_t num_tokens = input.numel() / input.size(-1); \
61
+ dim3 grid(num_tokens); \
62
+ dim3 block(std::min(d, 1024)); \
63
+ const at::cuda::OptionalCUDAGuard device_guard(device_of(input)); \
64
+ const cudaStream_t stream = at::cuda::getCurrentCUDAStream(); \
65
+ VLLM_DISPATCH_FLOATING_TYPES( \
66
+ input.scalar_type(), "act_and_mul_kernel", [&] { \
67
+ vllm::act_and_mul_kernel<scalar_t, KERNEL<scalar_t>> \
68
+ <<<grid, block, 0, stream>>>(out.data_ptr<scalar_t>(), \
69
+ input.data_ptr<scalar_t>(), d); \
70
+ });
71
+
72
+ void silu_and_mul(torch::Tensor& out, // [..., d]
73
+ torch::Tensor& input) // [..., 2 * d]
74
+ {
75
+ LAUNCH_ACTIVATION_GATE_KERNEL(vllm::silu_kernel);
76
+ }
77
+
78
+ void gelu_and_mul(torch::Tensor& out, // [..., d]
79
+ torch::Tensor& input) // [..., 2 * d]
80
+ {
81
+ LAUNCH_ACTIVATION_GATE_KERNEL(vllm::gelu_kernel);
82
+ }
83
+
84
+ void gelu_tanh_and_mul(torch::Tensor& out, // [..., d]
85
+ torch::Tensor& input) // [..., 2 * d]
86
+ {
87
+ LAUNCH_ACTIVATION_GATE_KERNEL(vllm::gelu_tanh_kernel);
88
+ }
89
+
90
+ namespace vllm {
91
+
92
+ template <typename T>
93
+ __device__ __forceinline__ T fatrelu_kernel(const T& x, const float threshold) {
94
+ const float f = (float)x;
95
+ return (T)(f > threshold ? f : 0.0f);
96
+ }
97
+
98
+ template <typename scalar_t, scalar_t (*ACT_FN)(const scalar_t&, const float)>
99
+ __global__ void act_and_mul_kernel_with_param(
100
+ scalar_t* __restrict__ out, const scalar_t* __restrict__ input, const int d,
101
+ const float param) {
102
+ const int64_t token_idx = blockIdx.x;
103
+ for (int64_t idx = threadIdx.x; idx < d; idx += blockDim.x) {
104
+ const scalar_t x = VLLM_LDG(&input[token_idx * 2 * d + idx]);
105
+ const scalar_t y = VLLM_LDG(&input[token_idx * 2 * d + d + idx]);
106
+ out[token_idx * d + idx] = ACT_FN(x, param) * y;
107
+ }
108
+ }
109
+
110
+ } // namespace vllm
111
+
112
+ #define LAUNCH_ACTIVATION_GATE_KERNEL_WITH_PARAM(KERNEL, PARAM) \
113
+ int d = input.size(-1) / 2; \
114
+ int64_t num_tokens = input.numel() / input.size(-1); \
115
+ dim3 grid(num_tokens); \
116
+ dim3 block(std::min(d, 1024)); \
117
+ const at::cuda::OptionalCUDAGuard device_guard(device_of(input)); \
118
+ const cudaStream_t stream = at::cuda::getCurrentCUDAStream(); \
119
+ VLLM_DISPATCH_FLOATING_TYPES( \
120
+ input.scalar_type(), "act_and_mul_kernel_with_param", [&] { \
121
+ vllm::act_and_mul_kernel_with_param<scalar_t, KERNEL<scalar_t>> \
122
+ <<<grid, block, 0, stream>>>(out.data_ptr<scalar_t>(), \
123
+ input.data_ptr<scalar_t>(), d, \
124
+ PARAM); \
125
+ });
126
+
127
+ void fatrelu_and_mul(torch::Tensor& out, // [..., d],
128
+ torch::Tensor& input, // [..., 2 * d]
129
+ double threshold) {
130
+ LAUNCH_ACTIVATION_GATE_KERNEL_WITH_PARAM(vllm::fatrelu_kernel, threshold);
131
+ }
132
+ namespace vllm {
133
+
134
+ // Element-wise activation kernel template.
135
+ template <typename scalar_t, scalar_t (*ACT_FN)(const scalar_t&)>
136
+ __global__ void activation_kernel(
137
+ scalar_t* __restrict__ out, // [..., d]
138
+ const scalar_t* __restrict__ input, // [..., d]
139
+ const int d) {
140
+ const int64_t token_idx = blockIdx.x;
141
+ for (int64_t idx = threadIdx.x; idx < d; idx += blockDim.x) {
142
+ const scalar_t x = VLLM_LDG(&input[token_idx * d + idx]);
143
+ out[token_idx * d + idx] = ACT_FN(x);
144
+ }
145
+ }
146
+
147
+ } // namespace vllm
148
+
149
+ // Launch element-wise activation kernel.
150
+ #define LAUNCH_ACTIVATION_KERNEL(KERNEL) \
151
+ int d = input.size(-1); \
152
+ int64_t num_tokens = input.numel() / d; \
153
+ dim3 grid(num_tokens); \
154
+ dim3 block(std::min(d, 1024)); \
155
+ const at::cuda::OptionalCUDAGuard device_guard(device_of(input)); \
156
+ const cudaStream_t stream = at::cuda::getCurrentCUDAStream(); \
157
+ VLLM_DISPATCH_FLOATING_TYPES(input.scalar_type(), "activation_kernel", [&] { \
158
+ vllm::activation_kernel<scalar_t, KERNEL<scalar_t>> \
159
+ <<<grid, block, 0, stream>>>(out.data_ptr<scalar_t>(), \
160
+ input.data_ptr<scalar_t>(), d); \
161
+ });
162
+
163
+ namespace vllm {
164
+
165
+ template <typename T>
166
+ __device__ __forceinline__ T gelu_new_kernel(const T& x) {
167
+ const float x3 = (float)(x * x * x);
168
+ const T t = (T)tanhf((T)(0.79788456f * (float)(x + (T)(0.044715f * x3))));
169
+ return ((T)0.5) * x * (((T)1.0) + t);
170
+ }
171
+
172
+ template <typename T>
173
+ __device__ __forceinline__ T gelu_fast_kernel(const T& x) {
174
+ const float f = (float)x;
175
+ const T t =
176
+ (T)tanhf(((T)(f * 0.79788456f)) * (((T)1.0) + (T)(0.044715f * f) * x));
177
+ return ((T)0.5) * x * (((T)1.0) + t);
178
+ }
179
+
180
+ template <typename T>
181
+ __device__ __forceinline__ T gelu_quick_kernel(const T& x) {
182
+ // x * sigmoid(1.702 * x)
183
+ return (T)(((float)x) / (1.0f + expf(-1.702f * (float)x)));
184
+ }
185
+
186
+ } // namespace vllm
187
+
188
+ void gelu_new(torch::Tensor& out, // [..., d]
189
+ torch::Tensor& input) // [..., d]
190
+ {
191
+ LAUNCH_ACTIVATION_KERNEL(vllm::gelu_new_kernel);
192
+ }
193
+
194
+ void gelu_fast(torch::Tensor& out, // [..., d]
195
+ torch::Tensor& input) // [..., d]
196
+ {
197
+ LAUNCH_ACTIVATION_KERNEL(vllm::gelu_fast_kernel);
198
+ }
199
+
200
+ void gelu_quick(torch::Tensor& out, // [..., d]
201
+ torch::Tensor& input) // [..., d]
202
+ {
203
+ LAUNCH_ACTIVATION_KERNEL(vllm::gelu_quick_kernel);
204
+ }
activation/cuda_compat.h ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #pragma once
2
+
3
+ #ifdef USE_ROCM
4
+ #include <hip/hip_runtime.h>
5
+ #endif
6
+
7
+ #ifndef USE_ROCM
8
+ #define WARP_SIZE 32
9
+ #else
10
+ #define WARP_SIZE warpSize
11
+ #endif
12
+
13
+ #ifndef USE_ROCM
14
+ #define VLLM_LDG(arg) __ldg(arg)
15
+ #else
16
+ #define VLLM_LDG(arg) *(arg)
17
+ #endif
18
+
19
+ #ifndef USE_ROCM
20
+ #define VLLM_SHFL_XOR_SYNC(var, lane_mask) \
21
+ __shfl_xor_sync(uint32_t(-1), var, lane_mask)
22
+ #define VLLM_SHFL_XOR_SYNC_WIDTH(var, lane_mask, width) \
23
+ __shfl_xor_sync(uint32_t(-1), var, lane_mask, width)
24
+ #else
25
+ #define VLLM_SHFL_XOR_SYNC(var, lane_mask) __shfl_xor(var, lane_mask)
26
+ #define VLLM_SHFL_XOR_SYNC_WIDTH(var, lane_mask, width) \
27
+ __shfl_xor(var, lane_mask, width)
28
+ #endif
29
+
30
+ #ifndef USE_ROCM
31
+ #define VLLM_SHFL_SYNC(var, src_lane) __shfl_sync(uint32_t(-1), var, src_lane)
32
+ #else
33
+ #define VLLM_SHFL_SYNC(var, src_lane) __shfl(var, src_lane)
34
+ #endif
35
+
36
+ #ifndef USE_ROCM
37
+ #define VLLM_SHFL_DOWN_SYNC(var, lane_delta) \
38
+ __shfl_down_sync(uint32_t(-1), var, lane_delta)
39
+ #else
40
+ #define VLLM_SHFL_DOWN_SYNC(var, lane_delta) __shfl_down(var, lane_delta)
41
+ #endif
42
+
43
+ #ifndef USE_ROCM
44
+ #define VLLM_DevFuncAttribute_SET_MaxDynamicSharedMemorySize(FUNC, VAL) \
45
+ cudaFuncSetAttribute(FUNC, cudaFuncAttributeMaxDynamicSharedMemorySize, VAL)
46
+ #else
47
+ #define VLLM_DevFuncAttribute_SET_MaxDynamicSharedMemorySize(FUNC, VAL) \
48
+ hipFuncSetAttribute(FUNC, hipFuncAttributeMaxDynamicSharedMemorySize, VAL)
49
+ #endif
activation/dispatch_utils.h ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ /*
2
+ * Adapted from
3
+ * https://github.com/pytorch/pytorch/blob/v2.0.1/aten/src/ATen/Dispatch.h
4
+ */
5
+ #pragma once
6
+
7
+ #include <torch/all.h>
8
+
9
+ #define VLLM_DISPATCH_CASE_FLOATING_TYPES(...) \
10
+ AT_DISPATCH_CASE(at::ScalarType::Float, __VA_ARGS__) \
11
+ AT_DISPATCH_CASE(at::ScalarType::Half, __VA_ARGS__) \
12
+ AT_DISPATCH_CASE(at::ScalarType::BFloat16, __VA_ARGS__)
13
+
14
+ #define VLLM_DISPATCH_FLOATING_TYPES(TYPE, NAME, ...) \
15
+ AT_DISPATCH_SWITCH(TYPE, NAME, VLLM_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__))
16
+
17
+ #define VLLM_DISPATCH_CASE_FLOATING_AND_BYTE_TYPES(...) \
18
+ AT_DISPATCH_CASE(at::ScalarType::Float, __VA_ARGS__) \
19
+ AT_DISPATCH_CASE(at::ScalarType::Half, __VA_ARGS__) \
20
+ AT_DISPATCH_CASE(at::ScalarType::BFloat16, __VA_ARGS__) \
21
+ AT_DISPATCH_CASE(at::ScalarType::Byte, __VA_ARGS__)
22
+
23
+ #define VLLM_DISPATCH_FLOATING_AND_BYTE_TYPES(TYPE, NAME, ...) \
24
+ AT_DISPATCH_SWITCH(TYPE, NAME, \
25
+ VLLM_DISPATCH_CASE_FLOATING_AND_BYTE_TYPES(__VA_ARGS__))
26
+
27
+ #define VLLM_DISPATCH_CASE_INTEGRAL_TYPES(...) \
28
+ AT_DISPATCH_CASE(at::ScalarType::Byte, __VA_ARGS__) \
29
+ AT_DISPATCH_CASE(at::ScalarType::Char, __VA_ARGS__) \
30
+ AT_DISPATCH_CASE(at::ScalarType::Short, __VA_ARGS__) \
31
+ AT_DISPATCH_CASE(at::ScalarType::Int, __VA_ARGS__) \
32
+ AT_DISPATCH_CASE(at::ScalarType::Long, __VA_ARGS__)
33
+
34
+ #define VLLM_DISPATCH_INTEGRAL_TYPES(TYPE, NAME, ...) \
35
+ AT_DISPATCH_SWITCH(TYPE, NAME, VLLM_DISPATCH_CASE_INTEGRAL_TYPES(__VA_ARGS__))
build.toml ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [torch]
2
+ name = "activation"
3
+ src = [
4
+ "ext-torch/registration.h",
5
+ "ext-torch/torch_binding.cpp",
6
+ "ext-torch/torch_binding.h"
7
+ ]
8
+ pysrc = [
9
+ "ext-torch/__init__.py"
10
+ ]
11
+
12
+ [kernel.activation]
13
+ capabilities = [ "7.0", "7.2", "7.5", "8.0", "8.6", "8.7", "8.9", "9.0" ]
14
+ src = [
15
+ "activation/activation_kernels.cu",
16
+ "activation/cuda_compat.h",
17
+ "activation/dispatch_utils.h",
18
+ ]
ext-torch/__init__.py ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch
2
+
3
+ import activation._activation
4
+
5
+ def silu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
6
+ torch.ops._activation.silu_and_mul(out, x)
7
+
8
+
9
+ def gelu_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
10
+ torch.ops._activation.gelu_and_mul(out, x)
11
+
12
+
13
+ def gelu_tanh_and_mul(out: torch.Tensor, x: torch.Tensor) -> None:
14
+ torch.ops._activation.gelu_tanh_and_mul(out, x)
15
+
16
+
17
+ def fatrelu_and_mul(out: torch.Tensor,
18
+ x: torch.Tensor,
19
+ threshold: float = 0.0) -> None:
20
+ torch.ops._activation.fatrelu_and_mul(out, x, threshold)
21
+
22
+
23
+ def gelu_fast(out: torch.Tensor, x: torch.Tensor) -> None:
24
+ torch.ops._activation.gelu_fast(out, x)
25
+
26
+
27
+ def gelu_new(out: torch.Tensor, x: torch.Tensor) -> None:
28
+ torch.ops._activation.gelu_new(out, x)
29
+
30
+
31
+ def gelu_quick(out: torch.Tensor, x: torch.Tensor) -> None:
32
+ torch.ops._activation.gelu_quick(out, x)
ext-torch/registration.h ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #pragma once
2
+
3
+ #include <Python.h>
4
+
5
+ #define _CONCAT(A, B) A##B
6
+ #define CONCAT(A, B) _CONCAT(A, B)
7
+
8
+ #define _STRINGIFY(A) #A
9
+ #define STRINGIFY(A) _STRINGIFY(A)
10
+
11
+ // A version of the TORCH_LIBRARY macro that expands the NAME, i.e. so NAME
12
+ // could be a macro instead of a literal token.
13
+ #define TORCH_LIBRARY_EXPAND(NAME, MODULE) TORCH_LIBRARY(NAME, MODULE)
14
+
15
+ // A version of the TORCH_LIBRARY_IMPL macro that expands the NAME, i.e. so NAME
16
+ // could be a macro instead of a literal token.
17
+ #define TORCH_LIBRARY_IMPL_EXPAND(NAME, DEVICE, MODULE) \
18
+ TORCH_LIBRARY_IMPL(NAME, DEVICE, MODULE)
19
+
20
+ // REGISTER_EXTENSION allows the shared library to be loaded and initialized
21
+ // via python's import statement.
22
+ #define REGISTER_EXTENSION(NAME) \
23
+ PyMODINIT_FUNC CONCAT(PyInit_, NAME)() { \
24
+ static struct PyModuleDef module = {PyModuleDef_HEAD_INIT, \
25
+ STRINGIFY(NAME), nullptr, 0, nullptr}; \
26
+ return PyModule_Create(&module); \
27
+ }
ext-torch/torch_binding.cpp ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #include <torch/library.h>
2
+
3
+ #include "registration.h"
4
+ #include "torch_binding.h"
5
+
6
+ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
7
+ // Activation ops
8
+ // Activation function used in SwiGLU.
9
+ ops.def("silu_and_mul(Tensor! out, Tensor input) -> ()");
10
+ ops.impl("silu_and_mul", torch::kCUDA, &silu_and_mul);
11
+
12
+ // Activation function used in GeGLU with `none` approximation.
13
+ ops.def("gelu_and_mul(Tensor! out, Tensor input) -> ()");
14
+ ops.impl("gelu_and_mul", torch::kCUDA, &gelu_and_mul);
15
+
16
+ // Activation function used in GeGLU with `tanh` approximation.
17
+ ops.def("gelu_tanh_and_mul(Tensor! out, Tensor input) -> ()");
18
+ ops.impl("gelu_tanh_and_mul", torch::kCUDA, &gelu_tanh_and_mul);
19
+
20
+ // FATReLU implementation.
21
+ ops.def("fatrelu_and_mul(Tensor! out, Tensor input, float threshold) -> ()");
22
+ ops.impl("fatrelu_and_mul", torch::kCUDA, &fatrelu_and_mul);
23
+
24
+ // GELU implementation used in GPT-2.
25
+ ops.def("gelu_new(Tensor! out, Tensor input) -> ()");
26
+ ops.impl("gelu_new", torch::kCUDA, &gelu_new);
27
+
28
+ // Approximate GELU implementation.
29
+ ops.def("gelu_fast(Tensor! out, Tensor input) -> ()");
30
+ ops.impl("gelu_fast", torch::kCUDA, &gelu_fast);
31
+ }
32
+
33
+ REGISTER_EXTENSION(TORCH_EXTENSION_NAME)
ext-torch/torch_binding.h ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #pragma once
2
+
3
+ #include <torch/torch.h>
4
+
5
+ void silu_and_mul(torch::Tensor &out, torch::Tensor &input);
6
+
7
+ void gelu_and_mul(torch::Tensor &out, torch::Tensor &input);
8
+
9
+ void gelu_tanh_and_mul(torch::Tensor &out, torch::Tensor &input);
10
+
11
+ void fatrelu_and_mul(torch::Tensor &out, torch::Tensor &input,
12
+ double threshold);
13
+
14
+ void gelu_new(torch::Tensor &out, torch::Tensor &input);
15
+
16
+ void gelu_fast(torch::Tensor &out, torch::Tensor &input);
17
+
18
+ void gelu_quick(torch::Tensor &out, torch::Tensor &input);