Skip to content

Commit b3176a4

Browse files
GasoonjiaGasoonjia
committed
Update base for Update on "[slimtensor] Add from_etensor factory function for ETensor to SlimTensor conversion"
Add from_etensor() factory function that creates a SlimTensor from an ExecuTorch portable tensor (ETensor), copying data to a target device. Key features: - Handles int32_t to int64_t conversion for sizes/strides (ETensor uses int32_t, SlimTensor uses int64_t) - Supports CPU and CUDA target devices via storage()->copy_() - Preserves tensor strides (non-contiguous layouts) - Provides both reference and pointer overloads Differential Revision: [D90539554](https://our.internmc.facebook.com/intern/diff/D90539554/) [ghstack-poisoned]
2 parents 1d996fa + 944a436 commit b3176a4

71 files changed

Lines changed: 1136 additions & 898 deletions

File tree

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

‎.ci/scripts/export_model_artifact.sh‎

Lines changed: 0 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -106,10 +106,6 @@ case "$HF_MODEL" in
106106
PREPROCESSOR_OUTPUT=""
107107
;;
108108
nvidia/parakeet-tdt)
109-
if [ "$DEVICE" = "metal" ]; then
110-
echo "Error: Export for device 'metal' is not yet tested for model '$HF_MODEL'"
111-
exit 1
112-
fi
113109
MODEL_NAME="parakeet"
114110
TASK=""
115111
MAX_SEQ_LEN=""

‎.github/workflows/cuda.yml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -71,7 +71,7 @@ jobs:
7171
strategy:
7272
fail-fast: false
7373
matrix:
74-
model: [linear, add, add_mul, resnet18, conv1d, sdpa]
74+
model: [linear, add, add_mul, resnet18, conv1d, sdpa, mv2, mv3]
7575
with:
7676
timeout: 90
7777
runner: linux.g5.4xlarge.nvidia.gpu

‎.github/workflows/metal.yml‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -44,6 +44,8 @@ jobs:
4444
name: "whisper-small"
4545
- repo: "openai"
4646
name: "whisper-large-v3-turbo"
47+
- repo: "nvidia"
48+
name: "parakeet-tdt"
4749
quant:
4850
- "non-quantized"
4951
with:
@@ -92,6 +94,8 @@ jobs:
9294
name: "whisper-small"
9395
- repo: "openai"
9496
name: "whisper-large-v3-turbo"
97+
- repo: "nvidia"
98+
name: "parakeet-tdt"
9599
quant:
96100
- "non-quantized"
97101
with:

‎backends/aoti/common_shims_slim.cpp‎

Lines changed: 1 addition & 103 deletions
Original file line numberDiff line numberDiff line change
@@ -64,110 +64,8 @@ int32_t aoti_torch_layout_strided() {
6464
return 0;
6565
}
6666

67-
// ============================================================
68-
// Storage & Device Property Getters - Implementations
69-
// ============================================================
70-
71-
AOTITorchError aoti_torch_get_storage_offset(
72-
Tensor* tensor,
73-
int64_t* ret_storage_offset) {
74-
if (tensor == nullptr || ret_storage_offset == nullptr) {
75-
return Error::InvalidArgument;
76-
}
77-
*ret_storage_offset = tensor->storage_offset();
78-
return Error::Ok;
79-
}
80-
81-
AOTITorchError aoti_torch_get_storage_size(Tensor* tensor, int64_t* ret_size) {
82-
if (tensor == nullptr || ret_size == nullptr) {
83-
return Error::InvalidArgument;
84-
}
85-
*ret_size = static_cast<int64_t>(tensor->storage()->nbytes());
86-
return Error::Ok;
87-
}
88-
89-
AOTITorchError aoti_torch_get_device_type(
90-
Tensor* tensor,
91-
int32_t* ret_device_type) {
92-
if (tensor == nullptr || ret_device_type == nullptr) {
93-
return Error::InvalidArgument;
94-
}
95-
*ret_device_type = static_cast<int32_t>(tensor->device_type());
96-
return Error::Ok;
97-
}
98-
99-
AOTITorchError aoti_torch_get_device_index(
100-
Tensor* tensor,
101-
int32_t* ret_device_index) {
102-
if (tensor == nullptr || ret_device_index == nullptr) {
103-
return Error::InvalidArgument;
104-
}
105-
*ret_device_index = static_cast<int32_t>(tensor->device_index());
106-
return Error::Ok;
107-
}
108-
109-
// ============================================================
110-
// DType Constants - Implementations
111-
// ============================================================
112-
113-
int32_t aoti_torch_dtype_float32() {
114-
return 6; // ScalarType::Float
115-
}
116-
117-
int32_t aoti_torch_dtype_bfloat16() {
118-
return 15; // ScalarType::BFloat16
119-
}
120-
121-
int32_t aoti_torch_dtype_int64() {
122-
return 4; // ScalarType::Long
123-
}
124-
125-
int32_t aoti_torch_dtype_int32() {
126-
return 3; // ScalarType::Int
127-
}
128-
129-
int32_t aoti_torch_dtype_int16() {
130-
return 2; // ScalarType::Short
131-
}
132-
133-
int32_t aoti_torch_dtype_int8() {
134-
return 1; // ScalarType::Char
135-
}
136-
137-
int32_t aoti_torch_dtype_bool() {
138-
return 11; // ScalarType::Bool
139-
}
140-
141-
// ============================================================
142-
// Device Type Constants - Implementations
143-
// ============================================================
144-
145-
int32_t aoti_torch_device_type_cpu() {
146-
return 0; // DeviceType::CPU
147-
}
148-
149-
int32_t aoti_torch_device_type_cuda() {
150-
return 1; // DeviceType::CUDA
151-
}
152-
153-
// ============================================================
154-
// Grad Mode Functions - Implementations
155-
// ============================================================
156-
157-
bool aoti_torch_grad_mode_is_enabled() {
158-
// ExecuTorch doesn't support autograd
159-
return false;
160-
}
161-
162-
AOTITorchError aoti_torch_grad_mode_set_enabled(bool enabled) {
163-
if (enabled) {
164-
// ExecuTorch doesn't support autograd
165-
return Error::NotSupported;
166-
}
167-
return Error::Ok;
168-
}
169-
17067
} // extern "C"
68+
17169
} // namespace aoti
17270
} // namespace backends
17371
} // namespace executorch

‎backends/aoti/common_shims_slim.h‎

Lines changed: 1 addition & 42 deletions
Original file line numberDiff line numberDiff line change
@@ -50,49 +50,8 @@ aoti_torch_get_dim(Tensor* tensor, int64_t* ret_dim);
5050

5151
AOTI_SHIM_EXPORT int32_t aoti_torch_layout_strided();
5252

53-
// ============================================================
54-
// Storage & Device Property Getters - Declarations
55-
// ============================================================
56-
57-
AOTI_SHIM_EXPORT AOTITorchError
58-
aoti_torch_get_storage_offset(Tensor* tensor, int64_t* ret_storage_offset);
59-
60-
AOTI_SHIM_EXPORT AOTITorchError
61-
aoti_torch_get_storage_size(Tensor* tensor, int64_t* ret_size);
62-
63-
AOTI_SHIM_EXPORT AOTITorchError
64-
aoti_torch_get_device_type(Tensor* tensor, int32_t* ret_device_type);
65-
66-
AOTI_SHIM_EXPORT AOTITorchError
67-
aoti_torch_get_device_index(Tensor* tensor, int32_t* ret_device_index);
68-
69-
// ============================================================
70-
// DType Constants - Declarations
71-
// ============================================================
72-
73-
AOTI_SHIM_EXPORT int32_t aoti_torch_dtype_float32();
74-
AOTI_SHIM_EXPORT int32_t aoti_torch_dtype_bfloat16();
75-
AOTI_SHIM_EXPORT int32_t aoti_torch_dtype_int64();
76-
AOTI_SHIM_EXPORT int32_t aoti_torch_dtype_int32();
77-
AOTI_SHIM_EXPORT int32_t aoti_torch_dtype_int16();
78-
AOTI_SHIM_EXPORT int32_t aoti_torch_dtype_int8();
79-
AOTI_SHIM_EXPORT int32_t aoti_torch_dtype_bool();
80-
81-
// ============================================================
82-
// Device Type Constants - Declarations
83-
// ============================================================
84-
85-
AOTI_SHIM_EXPORT int32_t aoti_torch_device_type_cpu();
86-
AOTI_SHIM_EXPORT int32_t aoti_torch_device_type_cuda();
87-
88-
// ============================================================
89-
// Grad Mode Functions - Declarations
90-
// ============================================================
91-
92-
AOTI_SHIM_EXPORT bool aoti_torch_grad_mode_is_enabled();
93-
AOTI_SHIM_EXPORT AOTITorchError aoti_torch_grad_mode_set_enabled(bool enabled);
94-
9553
} // extern "C"
54+
9655
} // namespace aoti
9756
} // namespace backends
9857
} // namespace executorch

‎backends/aoti/slim/core/test/BUCK‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,4 @@
1+
load("@fbcode_macros//build_defs:build_file_migration.bzl", "fbcode_target", "non_fbcode_target")
2+
load("targets.bzl", "define_common_targets")
3+
4+
fbcode_target(_kind = define_common_targets,)

‎backends/aoti/slim/core/test/TARGETS‎

Lines changed: 0 additions & 3 deletions
This file was deleted.

‎backends/aoti/slim/core/test/test_slimtensor_copy.cpp‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -10,6 +10,7 @@
1010

1111
#include <executorch/backends/aoti/slim/core/slim_tensor.h>
1212
#include <executorch/backends/aoti/slim/core/storage.h>
13+
#include <executorch/backends/aoti/slim/factory/empty.h>
1314

1415
namespace executorch::backends::aoti::slim {
1516

‎backends/aoti/slim/cuda/BUCK‎

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,7 @@
1+
load("@fbcode_macros//build_defs:build_file_migration.bzl", "fbcode_target", "non_fbcode_target")
2+
load("@fbsource//xplat/executorch/build:runtime_wrapper.bzl", "runtime")
3+
load(":targets.bzl", "define_common_targets")
4+
5+
oncall("executorch")
6+
7+
fbcode_target(_kind = define_common_targets,)

‎backends/aoti/slim/cuda/TARGETS‎

Lines changed: 0 additions & 6 deletions
This file was deleted.

0 commit comments

Comments
 (0)