Same code but different loss on Intel vs ARM

[berndporr]

🐛 Describe the bug

I've been trying to speed up transfer learning with mobilenetv2 by running the feature detector in executorch. The rest is libtorch. It all works nicely on my Intel PC with the loss converging to zero.
However, the same program on my Raspberry PI has a huge loss. It certainly also decays but about an order of magnitude higher:

https://github.com/berndporr/mobilenet_libtorch/tree/main/experimental

Executorch: v1.0.1 on both platforms.
Python part of executorch won't install on the RPI so ran the quantisation/export on the Intel PC.

Versions

Collecting environment information...
PyTorch version: 2.11.0.dev20251222+cpu
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: Ubuntu 24.04.3 LTS (x86_64)
GCC version: (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
Clang version: 18.1.3 (1ubuntu1)
CMake version: version 3.31.10
Libc version: glibc-2.39

Python version: 3.12.3 (main, Nov 6 2025, 13:44:16) [GCC 13.3.0] (64-bit runtime)
Python platform: Linux-6.14.0-37-generic-x86_64-with-glibc2.39
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
Is XPU available: False
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
Caching allocator config: N/A

CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 39 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 8
On-line CPU(s) list: 0-7
Vendor ID: GenuineIntel
Model name: Intel(R) Core(TM) i5-10210U CPU @ 1.60GHz
CPU family: 6
Model: 142
Thread(s) per core: 2
Core(s) per socket: 4
Socket(s): 1
Stepping: 12
CPU(s) scaling MHz: 42%
CPU max MHz: 4200.0000
CPU min MHz: 400.0000
BogoMIPS: 4199.88
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb ssbd ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust sgx bmi1 avx2 smep bmi2 erms invpcid mpx rdseed adx smap clflushopt intel_pt xsaveopt xsavec xgetbv1 xsaves dtherm ida arat pln pts hwp hwp_notify hwp_act_window hwp_epp vnmi md_clear flush_l1d arch_capabilities
Virtualisation: VT-x
L1d cache: 128 KiB (4 instances)
L1i cache: 128 KiB (4 instances)
L2 cache: 1 MiB (4 instances)
L3 cache: 6 MiB (1 instance)
NUMA node(s): 1
NUMA node0 CPU(s): 0-7
Vulnerability Gather data sampling: Vulnerable
Vulnerability Ghostwrite: Not affected
Vulnerability Indirect target selection: Mitigation; Aligned branch/return thunks
Vulnerability Itlb multihit: KVM: Mitigation: VMX disabled
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Mitigation; Clear CPU buffers; SMT vulnerable
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Mitigation; Enhanced IBRS
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI SW loop, KVM SW loop
Vulnerability Srbds: Mitigation; Microcode
Vulnerability Tsa: Not affected
Vulnerability Tsx async abort: Not affected
Vulnerability Vmscape: Mitigation; IBPB before exit to userspace

Versions of relevant libraries:
[pip3] executorch==1.1.0a0+5a64056
[pip3] numpy==2.3.4
[pip3] nvidia-cublas-cu12==12.8.4.1
[pip3] nvidia-cuda-cupti-cu12==12.8.90
[pip3] nvidia-cuda-nvrtc-cu12==12.8.93
[pip3] nvidia-cuda-runtime-cu12==12.8.90
[pip3] nvidia-cudnn-cu12==9.10.2.21
[pip3] nvidia-cufft-cu12==11.3.3.83
[pip3] nvidia-curand-cu12==10.3.9.90
[pip3] nvidia-cusolver-cu12==11.7.3.90
[pip3] nvidia-cusparse-cu12==12.5.8.93
[pip3] nvidia-cusparselt-cu12==0.7.1
[pip3] nvidia-nccl-cu12==2.27.5
[pip3] nvidia-nvjitlink-cu12==12.8.93
[pip3] nvidia-nvtx-cu12==12.8.90
[pip3] pytorch_tokenizers==1.0.1
[pip3] torch==2.11.0.dev20251222+cpu
[pip3] torchao==0.16.0+git08e5e203f
[pip3] torchaudio==2.10.0.dev20251222+cpu
[pip3] torchdata==0.11.0
[pip3] torchsr==1.0.4
[pip3] torchtune==0.6.1
[pip3] torchvision==0.25.0.dev20251222+cpu
[pip3] triton==3.5.1
[conda] Could not collect
(venv) bp1@ThinkPad-X1:~/sandbox/mobilenet_libtorch/experimental$

bp1@ThinkPad-X1:~/sandbox/executorch$ cmake --preset linux -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr/local -DEXECUTORCH_BUILD_EXTENSION_TRAINING=ON .
Preset CMake variables:

CMAKE_SYSTEM_NAME="Linux"
EXECUTORCH_BUILD_PRESET_FILE="/home/bp1/sandbox/executorch/tools/cmake/preset/linux.cmake"

-- Loading build preset: /home/bp1/sandbox/executorch/tools/cmake/preset/linux.cmake
-- ccache not found, builds will not be cached
-- Using the multi-header code from /home/bp1/sandbox/executorch/third-party/json/include/
CMake Deprecation Warning at third-party/gflags/CMakeLists.txt:73 (cmake_minimum_required):
Compatibility with CMake < 3.5 will be removed from a future version of
CMake.

Update the VERSION argument value or use a ... suffix to tell
CMake that the project does not need compatibility with older versions.

CMake Deprecation Warning at third-party/flatcc/CMakeLists.txt:2 (cmake_minimum_required):
Compatibility with CMake < 3.5 will be removed from a future version of
CMake.

Update the VERSION argument value or use a ... suffix to tell
CMake that the project does not need compatibility with older versions.

-- dist install dir /home/bp1/sandbox/executorch/third-party/flatcc
-- lib install dir /home/bp1/sandbox/executorch/third-party/flatcc/lib
-- Setting GNU C compiler options with c11 and Posix
-- Disabling -pedantic for GCC >= 8.0
-- Disabling GNU C compiler warnings: -Wstringop-truncation -Wno-format-overflow
-- GCC_VERSION: 13

-- Configured C_FLAGS: -DFLATCC_REFLECTION=0 -std=c11 -Wall -Wextra -Wno-stringop-truncation -Wno-format-overflow -Wno-misleading-indentation -DPORTABLE_POSIX_MEMALIGN=1 -Werror -Wno-unused-function -Wsign-conversion
'python3' '-c' 'import importlib.util; print(importlib.util.find_spec('torch').submodule_search_locations[0])'
Error while adding torch to CMAKE_PREFIX_PATH. Exit code: 1
Output:

CMake Error at tools/cmake/Utils.cmake:135 (message):
Error:

Traceback (most recent call last):

File "<string>", line 1, in <module>

AttributeError: 'NoneType' object has no attribute
'submodule_search_locations'

Call Stack (most recent call first):
tools/cmake/Utils.cmake:109 (get_torch_base_path)
CMakeLists.txt:482 (find_package_torch_headers)

-- Configuring incomplete, errors occurred!
bp1@ThinkPad-X1:~/sandbox/executorch$ source /venv/bin/activate
(venv) bp1@ThinkPad-X1:/sandbox/executorch$ cmake --preset linux -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr/local -DEXECUTORCH_BUILD_EXTENSION_TRAINING=ON .
Preset CMake variables:

CMAKE_SYSTEM_NAME="Linux"
EXECUTORCH_BUILD_PRESET_FILE="/home/bp1/sandbox/executorch/tools/cmake/preset/linux.cmake"

-- Loading build preset: /home/bp1/sandbox/executorch/tools/cmake/preset/linux.cmake
-- ccache not found, builds will not be cached
-- Using the multi-header code from /home/bp1/sandbox/executorch/third-party/json/include/
CMake Deprecation Warning at third-party/gflags/CMakeLists.txt:73 (cmake_minimum_required):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.

Update the VERSION argument value. Or, use the ... syntax
to tell CMake that the project requires at least but has been updated
to work with policies introduced by or earlier.

CMake Deprecation Warning at third-party/flatcc/CMakeLists.txt:2 (cmake_minimum_required):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.

Update the VERSION argument value. Or, use the ... syntax
to tell CMake that the project requires at least but has been updated
to work with policies introduced by or earlier.

-- dist install dir /home/bp1/sandbox/executorch/third-party/flatcc
-- lib install dir /home/bp1/sandbox/executorch/third-party/flatcc/lib
-- Setting GNU C compiler options with c11 and Posix
-- Disabling -pedantic for GCC >= 8.0
-- Disabling GNU C compiler warnings: -Wstringop-truncation -Wno-format-overflow
-- GCC_VERSION: 13

-- Configured C_FLAGS: -DFLATCC_REFLECTION=0 -std=c11 -Wall -Wextra -Wno-stringop-truncation -Wno-format-overflow -Wno-misleading-indentation -DPORTABLE_POSIX_MEMALIGN=1 -Werror -Wno-unused-function -Wsign-conversion
CMake Deprecation Warning at backends/xnnpack/third-party/FXdiv/CMakeLists.txt:1 (CMAKE_MINIMUM_REQUIRED):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.

Update the VERSION argument value. Or, use the ... syntax
to tell CMake that the project requires at least but has been updated
to work with policies introduced by or earlier.

CMake Deprecation Warning at backends/xnnpack/third-party/cpuinfo/CMakeLists.txt:1 (CMAKE_MINIMUM_REQUIRED):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.

Update the VERSION argument value. Or, use the ... syntax
to tell CMake that the project requires at least but has been updated
to work with policies introduced by or earlier.

CMake Deprecation Warning at backends/xnnpack/third-party/pthreadpool/CMakeLists.txt:10 (CMAKE_MINIMUM_REQUIRED):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.

Update the VERSION argument value. Or, use the ... syntax
to tell CMake that the project requires at least but has been updated
to work with policies introduced by or earlier.

'/home/bp1/venv/bin/python3' '-c' 'import importlib.util; print(importlib.util.find_spec('torch').submodule_search_locations[0])'
-- Generating selected operator lib:
-- LIB_NAME: portable_ops_lib
-- OPS_SCHEMA_YAML: /home/bp1/sandbox/executorch/kernels/portable/functions.yaml
-- ROOT_OPS:
-- INCLUDE_ALL_OPS:
-- OPS_FROM_MODEL:
-- DTYPE_SELECTIVE_BUILD:
Command - /home/bp1/venv/bin/python3;-m;codegen.tools.gen_oplist;--output_path=/home/bp1/sandbox/executorch/cmake-out/kernels/portable/portable_ops_lib/selected_operators.yaml;--ops_schema_yaml_path="/home/bp1/sandbox/executorch/kernels/portable/functions.yaml"
-- Generating kernel bindings:
-- LIB_NAME: portable_ops_lib
-- FUNCTIONS_YAML: /home/bp1/sandbox/executorch/kernels/portable/functions.yaml
-- CUSTOM_OPS_YAML:
-- ADD_EXCEPTION_BOUNDARY: FALSE
-- DTYPE_SELECTIVE_BUILD:
Generated files /home/bp1/sandbox/executorch/cmake-out/kernels/portable/portable_ops_lib/RegisterCodegenUnboxedKernelsEverything.cpp;/home/bp1/sandbox/executorch/cmake-out/kernels/portable/portable_ops_lib/Functions.h;/home/bp1/sandbox/executorch/cmake-out/kernels/portable/portable_ops_lib/NativeFunctions.h
-- Generating operator lib:
-- LIB_NAME: portable_ops_lib
-- KERNEL_LIBS: portable_kernels
-- DEPS: executorch_core
-- DTYPE_SELECTIVE_BUILD:
-- Generating selected operator lib:
-- LIB_NAME: optimized_portable_ops_lib
-- OPS_SCHEMA_YAML: /home/bp1/sandbox/executorch/kernels/portable/functions.yaml
-- ROOT_OPS:
-- INCLUDE_ALL_OPS:
-- OPS_FROM_MODEL:
-- DTYPE_SELECTIVE_BUILD:
Command - /home/bp1/venv/bin/python3;-m;codegen.tools.gen_oplist;--output_path=/home/bp1/sandbox/executorch/cmake-out/kernels/portable/optimized_portable_ops_lib/selected_operators.yaml;--ops_schema_yaml_path="/home/bp1/sandbox/executorch/kernels/portable/functions.yaml"
-- Generating kernel bindings:
-- LIB_NAME: optimized_portable_ops_lib
-- FUNCTIONS_YAML: /home/bp1/sandbox/executorch/kernels/portable/functions.yaml
-- CUSTOM_OPS_YAML:
-- ADD_EXCEPTION_BOUNDARY: FALSE
-- DTYPE_SELECTIVE_BUILD:
-- Generating operator lib:
-- LIB_NAME: optimized_portable_ops_lib
-- KERNEL_LIBS: optimized_portable_kernels
-- DEPS: executorch_core
-- DTYPE_SELECTIVE_BUILD:
-- Generating selected operator lib:
-- LIB_NAME: optimized_ops_lib
-- OPS_SCHEMA_YAML: /home/bp1/sandbox/executorch/kernels/optimized/optimized.yaml
-- ROOT_OPS:
-- INCLUDE_ALL_OPS:
-- OPS_FROM_MODEL:
-- DTYPE_SELECTIVE_BUILD:
Command - /home/bp1/venv/bin/python3;-m;codegen.tools.gen_oplist;--output_path=/home/bp1/sandbox/executorch/cmake-out/kernels/optimized/optimized_ops_lib/selected_operators.yaml;--ops_schema_yaml_path="/home/bp1/sandbox/executorch/kernels/optimized/optimized.yaml"
-- Generating kernel bindings:
-- LIB_NAME: optimized_ops_lib
-- FUNCTIONS_YAML: /home/bp1/sandbox/executorch/kernels/optimized/optimized.yaml
-- CUSTOM_OPS_YAML:
-- ADD_EXCEPTION_BOUNDARY: TRUE
-- DTYPE_SELECTIVE_BUILD:
Generated files /home/bp1/sandbox/executorch/cmake-out/kernels/optimized/optimized_ops_lib/RegisterCodegenUnboxedKernelsEverything.cpp;/home/bp1/sandbox/executorch/cmake-out/kernels/optimized/optimized_ops_lib/Functions.h;/home/bp1/sandbox/executorch/cmake-out/kernels/optimized/optimized_ops_lib/NativeFunctions.h
-- Generating operator lib:
-- LIB_NAME: optimized_ops_lib
-- KERNEL_LIBS: optimized_kernels
-- DEPS: executorch_core
-- DTYPE_SELECTIVE_BUILD:
-- Merging kernel yaml files:
-- FUNCTIONS_YAML: /home/bp1/sandbox/executorch/kernels/optimized/optimized.yaml
-- FALLBACK_YAML: /home/bp1/sandbox/executorch/kernels/portable/functions.yaml
-- OUTPUT_DIR: /home/bp1/sandbox/executorch/cmake-out/configurations
-- Generating selected operator lib:
-- LIB_NAME: optimized_native_cpu_ops_lib
-- OPS_SCHEMA_YAML: /home/bp1/sandbox/executorch/cmake-out/configurations/merged.yaml
-- ROOT_OPS:
-- INCLUDE_ALL_OPS:
-- OPS_FROM_MODEL:
-- DTYPE_SELECTIVE_BUILD:
Command - /home/bp1/venv/bin/python3;-m;codegen.tools.gen_oplist;--output_path=/home/bp1/sandbox/executorch/cmake-out/configurations/optimized_native_cpu_ops_lib/selected_operators.yaml;--ops_schema_yaml_path="/home/bp1/sandbox/executorch/cmake-out/configurations/merged.yaml"
-- Generating kernel bindings:
-- LIB_NAME: optimized_native_cpu_ops_lib
-- FUNCTIONS_YAML: /home/bp1/sandbox/executorch/cmake-out/configurations/merged.yaml
-- CUSTOM_OPS_YAML:
-- ADD_EXCEPTION_BOUNDARY: TRUE
-- DTYPE_SELECTIVE_BUILD:
Generated files /home/bp1/sandbox/executorch/cmake-out/configurations/optimized_native_cpu_ops_lib/RegisterCodegenUnboxedKernelsEverything.cpp;/home/bp1/sandbox/executorch/cmake-out/configurations/optimized_native_cpu_ops_lib/Functions.h;/home/bp1/sandbox/executorch/cmake-out/configurations/optimized_native_cpu_ops_lib/NativeFunctions.h
-- Generating operator lib:
-- LIB_NAME: optimized_native_cpu_ops_lib
-- KERNEL_LIBS: optimized_kernels;optimized_portable_kernels
-- DEPS: executorch_core
-- DTYPE_SELECTIVE_BUILD:
-- Building for XNNPACK_TARGET_PROCESSOR: x86_64
-- Generating microkernels.cmake
CMake Warning at extension/llm/tokenizers/third-party/abseil-cpp/CMakeLists.txt:177 (message):
The default and system-level install directories are unsupported except in LTS releases of Abseil. Please set CMAKE_INSTALL_PREFIX to install Abseil in your source or build tree directly.

CMake Deprecation Warning at extension/llm/tokenizers/third-party/sentencepiece/CMakeLists.txt:15 (cmake_minimum_required):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.

Update the VERSION argument value. Or, use the ... syntax
to tell CMake that the project requires at least but has been updated
to work with policies introduced by or earlier.

-- VERSION: 0.2.1
-- Not Found TCMalloc: TCMALLOC_LIB-NOTFOUND
-- Could NOT find BZip2 (missing: BZIP2_LIBRARIES BZIP2_INCLUDE_DIR)
-- Could NOT find Readline (missing: READLINE_INCLUDE_DIR READLINE_LIBRARY)
-- Could NOT find Editline (missing: EDITLINE_INCLUDE_DIR EDITLINE_LIBRARY)

--
-- PCRE2-10.46 configuration summary:

-- Install prefix .................... : /usr/local
-- C compiler ........................ : /usr/bin/cc
-- C compiler flags .................. : -mshstk -O3 -DNDEBUG

-- Build type ........................ : Release
-- Build 8 bit pcre2 library ......... : ON
-- Build 16 bit pcre2 library ........ : OFF
-- Build 32 bit pcre2 library ........ : OFF
-- Include debugging code ............ : IfDebugBuild
-- Enable JIT compiling support ...... : OFF
-- Use SELinux allocator in JIT ...... : OFF
-- Enable Unicode support ............ : ON
-- Newline char/sequence ............. : LF
-- \R matches only ANYCRLF ........... : OFF
-- \C is disabled .................... : OFF
-- EBCDIC coding ..................... : OFF
-- EBCDIC code for NL ................ : n/a
-- EBCDIC coding ignoring compiler ... : OFF
-- Rebuild char tables ............... : OFF
-- Internal link size ................ : 2
-- Maximum variable lookbehind ....... : 255
-- Nested parentheses limit .......... : 250
-- Heap limit ........................ : 20000000
-- Match limit ....................... : 10000000
-- Match depth limit ................. : MATCH_LIMIT
-- Build shared libs ................. : OFF
-- with symbol versioning ........ : ON
-- Build static libs ................. : ON
-- with PIC enabled .............. : ON
-- Build pcre2grep ................... : OFF
-- Enable JIT in pcre2grep ........... : ON
-- Enable callouts in pcre2grep ...... : ON
-- Enable callout fork in pcre2grep .. : ON
-- Initial buffer size for pcre2grep . : 20480
-- Maximum buffer size for pcre2grep . : 1048576
-- Build tests (implies pcre2test .... : OFF
-- and pcre2grep)
-- Link pcre2grep with libz .......... : ON
-- Link pcre2grep with libbz2 ........ : Library not found
-- Link pcre2test with libeditline ... : Library not found
-- Link pcre2test with libreadline ... : Library not found
-- Enable Valgrind support ........... : OFF
-- Use %zu and %td ................... : AUTO

CMake Warning (dev) at extension/llm/tokenizers/CMakeLists.txt:204 (install):
Target re2 has PUBLIC_HEADER files but no PUBLIC_HEADER DESTINATION.
This warning is for project developers. Use -Wno-dev to suppress it.

'/home/bp1/venv/bin/python3' '-c' 'import importlib.util; print(importlib.util.find_spec('torch').submodule_search_locations[0])'
-- Generating selected operator lib:
-- LIB_NAME: quantized_ops_lib
-- OPS_SCHEMA_YAML: /home/bp1/sandbox/executorch/kernels/quantized/quantized.yaml
-- ROOT_OPS:
-- INCLUDE_ALL_OPS:
-- OPS_FROM_MODEL:
-- DTYPE_SELECTIVE_BUILD:
Command - /home/bp1/venv/bin/python3;-m;codegen.tools.gen_oplist;--output_path=/home/bp1/sandbox/executorch/cmake-out/kernels/quantized/quantized_ops_lib/selected_operators.yaml;--ops_schema_yaml_path="/home/bp1/sandbox/executorch/kernels/quantized/quantized.yaml"
-- Generating kernel bindings:
-- LIB_NAME: quantized_ops_lib
-- FUNCTIONS_YAML:
-- CUSTOM_OPS_YAML: /home/bp1/sandbox/executorch/kernels/quantized/quantized.yaml
-- ADD_EXCEPTION_BOUNDARY: FALSE
-- DTYPE_SELECTIVE_BUILD:
Generated files /home/bp1/sandbox/executorch/cmake-out/kernels/quantized/quantized_ops_lib/RegisterCodegenUnboxedKernelsEverything.cpp;/home/bp1/sandbox/executorch/cmake-out/kernels/quantized/quantized_ops_lib/Functions.h;/home/bp1/sandbox/executorch/cmake-out/kernels/quantized/quantized_ops_lib/NativeFunctions.h;/home/bp1/sandbox/executorch/cmake-out/kernels/quantized/quantized_ops_lib/RegisterCPUCustomOps.cpp;/home/bp1/sandbox/executorch/cmake-out/kernels/quantized/quantized_ops_lib/RegisterSchema.cpp;/home/bp1/sandbox/executorch/cmake-out/kernels/quantized/quantized_ops_lib/CustomOpsNativeFunctions.h
-- Generating operator lib:
-- LIB_NAME: quantized_ops_lib
-- KERNEL_LIBS: quantized_kernels
-- DEPS: executorch_core
-- DTYPE_SELECTIVE_BUILD:
-- --- Configured Options ---

-- BUILD_TESTING : OFF
-- CCACHE_PROGRAM : CCACHE_PROGRAM-NOTFOUND
-- CMAKE_BUILD_TYPE : Release
-- CMAKE_CXX_COMPILER_ID : GNU
-- CMAKE_CXX_STANDARD : 17
-- CMAKE_SYSTEM_PROCESSOR : x86_64
-- CMAKE_TOOLCHAIN_FILE x (unset)
-- EXECUTORCH_BUILD_ARM_BAREMETAL : OFF
-- EXECUTORCH_BUILD_CADENCE : OFF
-- EXECUTORCH_BUILD_COREML : OFF
-- EXECUTORCH_BUILD_CORTEX_M : OFF
-- EXECUTORCH_BUILD_CPUINFO : ON
-- EXECUTORCH_BUILD_DEVTOOLS : OFF
-- EXECUTORCH_BUILD_EXECUTOR_RUNNER : ON
-- EXECUTORCH_BUILD_EXTENSION_APPLE : OFF
-- EXECUTORCH_BUILD_EXTENSION_DATA_LOADER : ON
-- EXECUTORCH_BUILD_EXTENSION_EVALUE_UTIL : ON
-- EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR : ON
-- EXECUTORCH_BUILD_EXTENSION_LLM : ON
-- EXECUTORCH_BUILD_EXTENSION_LLM_APPLE : OFF
-- EXECUTORCH_BUILD_EXTENSION_LLM_RUNNER : ON
-- EXECUTORCH_BUILD_EXTENSION_MODULE : ON
-- EXECUTORCH_BUILD_EXTENSION_NAMED_DATA_MAP : ON
-- EXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL : ON
-- EXECUTORCH_BUILD_EXTENSION_TENSOR : ON
-- EXECUTORCH_BUILD_EXTENSION_TRAINING : ON
-- EXECUTORCH_BUILD_KERNELS_LLM : ON
-- EXECUTORCH_BUILD_KERNELS_LLM_AOT : OFF
-- EXECUTORCH_BUILD_KERNELS_OPTIMIZED : ON
-- EXECUTORCH_BUILD_KERNELS_QUANTIZED : ON
-- EXECUTORCH_BUILD_KERNELS_QUANTIZED_AOT : OFF
-- EXECUTORCH_BUILD_MPS : OFF
-- EXECUTORCH_BUILD_NEURON : OFF
-- EXECUTORCH_BUILD_OPENVINO : OFF
-- EXECUTORCH_BUILD_PORTABLE_OPS : ON
-- EXECUTORCH_BUILD_PRESET_FILE : /home/bp1/sandbox/executorch/tools/cmake/preset/linux.cmake
-- EXECUTORCH_BUILD_PTHREADPOOL : ON
-- EXECUTORCH_BUILD_PYBIND : OFF
-- EXECUTORCH_BUILD_QNN : OFF
-- EXECUTORCH_BUILD_SIZE_TEST : OFF
-- EXECUTORCH_BUILD_TESTS : OFF
-- EXECUTORCH_BUILD_TOKENIZERS_WASM : OFF
-- EXECUTORCH_BUILD_VGF : OFF
-- EXECUTORCH_BUILD_VULKAN : OFF
-- EXECUTORCH_BUILD_WASM : OFF
-- EXECUTORCH_BUILD_XNNPACK : ON
-- EXECUTORCH_CADENCE_CPU_RUNNER : OFF
-- EXECUTORCH_COREML_BUILD_EXECUTOR_RUNNER : OFF
-- EXECUTORCH_ENABLE_DTYPE_SELECTIVE_BUILD : FALSE
-- EXECUTORCH_ENABLE_EVENT_TRACER : OFF
-- EXECUTORCH_ENABLE_LOGGING : OFF
-- EXECUTORCH_ENABLE_PROGRAM_VERIFICATION : OFF
-- EXECUTORCH_FLATBUFFERS_MAX_ALIGNMENT : 1024
-- EXECUTORCH_LOG_LEVEL : Info
-- EXECUTORCH_NNLIB_OPT : OFF
-- EXECUTORCH_OPTIMIZE_SIZE : OFF
-- EXECUTORCH_PAL_DEFAULT : posix
-- EXECUTORCH_PAL_DEFAULT_FILE_PATH : /home/bp1/sandbox/executorch/runtime/platform/default/posix.cpp
-- EXECUTORCH_SELECT_OPS_LIST :
-- EXECUTORCH_SELECT_OPS_MODEL :
-- EXECUTORCH_SELECT_OPS_YAML :
-- EXECUTORCH_USE_CPP_CODE_COVERAGE : OFF
-- EXECUTORCH_USE_DL : ON
-- EXECUTORCH_XNNPACK_ENABLE_KLEIDI : ON
-- EXECUTORCH_XNNPACK_ENABLE_WEIGHT_CACHE : OFF
-- EXECUTORCH_XNNPACK_SHARED_WORKSPACE : ON
-- PYTHON_EXECUTABLE : /home/bp1/venv/bin/python3

---

-- Configuring done (7.9s)
-- Generating done (1.2s)
-- Build files have been written to: /home/bp1/sandbox/executorch/cmake-out
(venv) bp1@ThinkPad-X1:~/sandbox/executorch$

---

initialize NumPy: No module named 'numpy' (Triggered internally at /pytorch/torch/csrc/utils/tensor_numpy.cpp:84.)
cpu = _conversion_method_template(device=torch.device("cpu"))
Collecting environment information...
PyTorch version: 2.11.0.dev20251222+cpu
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A

OS: Debian GNU/Linux 13 (trixie) (aarch64)
GCC version: (Debian 14.2.0-19) 14.2.0
Clang version: 14.0.6
CMake version: version 3.31.10
Libc version: glibc-2.41

Python version: 3.13.5 (main, Jun 25 2025, 18:55:22) [GCC 14.2.0] (64-bit runtime)
Python platform: Linux-6.12.47+rpt-rpi-2712-aarch64-with-glibc2.41
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
Is XPU available: False
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
Caching allocator config: N/A

CPU:
Architecture: aarch64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 4
On-line CPU(s) list: 0-3
Vendor ID: ARM
Model name: Cortex-A76
Model: 1
Thread(s) per core: 1
Core(s) per cluster: 4
Socket(s): -
Cluster(s): 1
Stepping: r4p1
CPU(s) scaling MHz: 100%
CPU max MHz: 2400.0000
CPU min MHz: 1500.0000
BogoMIPS: 108.00
Flags: fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm lrcpc dcpop asimddp
L1d cache: 256 KiB (4 instances)
L1i cache: 256 KiB (4 instances)
L2 cache: 2 MiB (4 instances)
L3 cache: 2 MiB (1 instance)
NUMA node(s): 8
NUMA node0 CPU(s): 0-3
NUMA node1 CPU(s): 0-3
NUMA node2 CPU(s): 0-3
NUMA node3 CPU(s): 0-3
NUMA node4 CPU(s): 0-3
NUMA node5 CPU(s): 0-3
NUMA node6 CPU(s): 0-3
NUMA node7 CPU(s): 0-3
Vulnerability Gather data sampling: Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1: Mitigation; __user pointer sanitization
Vulnerability Spectre v2: Mitigation; CSV2, BHB
Vulnerability Srbds: Not affected
Vulnerability Tsa: Not affected
Vulnerability Tsx async abort: Not affected
Vulnerability Vmscape: Not affected

Versions of relevant libraries:
[pip3] torch==2.11.0.dev20251222+cpu
[conda] Could not collect

(venv) bp1@gngit:~/executorch $ cmake --preset linux -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr/local .
Preset CMake variables:

CMAKE_SYSTEM_NAME="Linux"
EXECUTORCH_BUILD_PRESET_FILE="/home/bp1/executorch/tools/cmake/preset/linux.cmake"

-- Loading build preset: /home/bp1/executorch/tools/cmake/preset/linux.cmake
-- ccache not found, builds will not be cached
-- Using the multi-header code from /home/bp1/executorch/third-party/json/include/
CMake Deprecation Warning at third-party/gflags/CMakeLists.txt:73 (cmake_minimum_required):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.

Update the VERSION argument value. Or, use the ... syntax
to tell CMake that the project requires at least but has been updated
to work with policies introduced by or earlier.

CMake Deprecation Warning at third-party/flatcc/CMakeLists.txt:2 (cmake_minimum_required):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.

Update the VERSION argument value. Or, use the ... syntax
to tell CMake that the project requires at least but has been updated
to work with policies introduced by or earlier.

-- dist install dir /home/bp1/executorch/third-party/flatcc
-- lib install dir /home/bp1/executorch/third-party/flatcc/lib
-- Setting GNU C compiler options with c11 and Posix
-- Disabling -pedantic for GCC >= 8.0
-- Disabling GNU C compiler warnings: -Wstringop-truncation -Wno-format-overflow
-- GCC_VERSION: 14

-- Configured C_FLAGS: -DFLATCC_REFLECTION=0 -std=c11 -Wall -Wextra -Wno-stringop-truncation -Wno-format-overflow -Wno-misleading-indentation -DPORTABLE_POSIX_MEMALIGN=1 -Werror -Wno-unused-function -Wsign-conversion
CMake Deprecation Warning at backends/xnnpack/third-party/FXdiv/CMakeLists.txt:1 (CMAKE_MINIMUM_REQUIRED):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.

Update the VERSION argument value. Or, use the ... syntax
to tell CMake that the project requires at least but has been updated
to work with policies introduced by or earlier.

CMake Deprecation Warning at backends/xnnpack/third-party/cpuinfo/CMakeLists.txt:1 (CMAKE_MINIMUM_REQUIRED):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.

Update the VERSION argument value. Or, use the ... syntax
to tell CMake that the project requires at least but has been updated
to work with policies introduced by or earlier.

CMake Deprecation Warning at backends/xnnpack/third-party/pthreadpool/CMakeLists.txt:10 (CMAKE_MINIMUM_REQUIRED):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.

Update the VERSION argument value. Or, use the ... syntax
to tell CMake that the project requires at least but has been updated
to work with policies introduced by or earlier.

'/home/bp1/venv/bin/python3' '-c' 'import importlib.util; print(importlib.util.find_spec('torch').submodule_search_locations[0])'
-- Generating selected operator lib:
-- LIB_NAME: portable_ops_lib
-- OPS_SCHEMA_YAML: /home/bp1/executorch/kernels/portable/functions.yaml
-- ROOT_OPS:
-- INCLUDE_ALL_OPS:
-- OPS_FROM_MODEL:
-- DTYPE_SELECTIVE_BUILD:
Command - /home/bp1/venv/bin/python3;-m;codegen.tools.gen_oplist;--output_path=/home/bp1/executorch/cmake-out/kernels/portable/portable_ops_lib/selected_operators.yaml;--ops_schema_yaml_path="/home/bp1/executorch/kernels/portable/functions.yaml"
-- Generating kernel bindings:
-- LIB_NAME: portable_ops_lib
-- FUNCTIONS_YAML: /home/bp1/executorch/kernels/portable/functions.yaml
-- CUSTOM_OPS_YAML:
-- ADD_EXCEPTION_BOUNDARY: FALSE
-- DTYPE_SELECTIVE_BUILD:
Generated files /home/bp1/executorch/cmake-out/kernels/portable/portable_ops_lib/RegisterCodegenUnboxedKernelsEverything.cpp;/home/bp1/executorch/cmake-out/kernels/portable/portable_ops_lib/Functions.h;/home/bp1/executorch/cmake-out/kernels/portable/portable_ops_lib/NativeFunctions.h
-- Generating operator lib:
-- LIB_NAME: portable_ops_lib
-- KERNEL_LIBS: portable_kernels
-- DEPS: executorch_core
-- DTYPE_SELECTIVE_BUILD:
-- Generating selected operator lib:
-- LIB_NAME: optimized_portable_ops_lib
-- OPS_SCHEMA_YAML: /home/bp1/executorch/kernels/portable/functions.yaml
-- ROOT_OPS:
-- INCLUDE_ALL_OPS:
-- OPS_FROM_MODEL:
-- DTYPE_SELECTIVE_BUILD:
Command - /home/bp1/venv/bin/python3;-m;codegen.tools.gen_oplist;--output_path=/home/bp1/executorch/cmake-out/kernels/portable/optimized_portable_ops_lib/selected_operators.yaml;--ops_schema_yaml_path="/home/bp1/executorch/kernels/portable/functions.yaml"
-- Generating kernel bindings:
-- LIB_NAME: optimized_portable_ops_lib
-- FUNCTIONS_YAML: /home/bp1/executorch/kernels/portable/functions.yaml
-- CUSTOM_OPS_YAML:
-- ADD_EXCEPTION_BOUNDARY: FALSE
-- DTYPE_SELECTIVE_BUILD:
-- Generating operator lib:
-- LIB_NAME: optimized_portable_ops_lib
-- KERNEL_LIBS: optimized_portable_kernels
-- DEPS: executorch_core
-- DTYPE_SELECTIVE_BUILD:
-- Generating selected operator lib:
-- LIB_NAME: optimized_ops_lib
-- OPS_SCHEMA_YAML: /home/bp1/executorch/kernels/optimized/optimized.yaml
-- ROOT_OPS:
-- INCLUDE_ALL_OPS:
-- OPS_FROM_MODEL:
-- DTYPE_SELECTIVE_BUILD:
Command - /home/bp1/venv/bin/python3;-m;codegen.tools.gen_oplist;--output_path=/home/bp1/executorch/cmake-out/kernels/optimized/optimized_ops_lib/selected_operators.yaml;--ops_schema_yaml_path="/home/bp1/executorch/kernels/optimized/optimized.yaml"
-- Generating kernel bindings:
-- LIB_NAME: optimized_ops_lib
-- FUNCTIONS_YAML: /home/bp1/executorch/kernels/optimized/optimized.yaml
-- CUSTOM_OPS_YAML:
-- ADD_EXCEPTION_BOUNDARY: TRUE
-- DTYPE_SELECTIVE_BUILD:
Generated files /home/bp1/executorch/cmake-out/kernels/optimized/optimized_ops_lib/RegisterCodegenUnboxedKernelsEverything.cpp;/home/bp1/executorch/cmake-out/kernels/optimized/optimized_ops_lib/Functions.h;/home/bp1/executorch/cmake-out/kernels/optimized/optimized_ops_lib/NativeFunctions.h
-- Generating operator lib:
-- LIB_NAME: optimized_ops_lib
-- KERNEL_LIBS: optimized_kernels
-- DEPS: executorch_core
-- DTYPE_SELECTIVE_BUILD:
-- Merging kernel yaml files:
-- FUNCTIONS_YAML: /home/bp1/executorch/kernels/optimized/optimized.yaml
-- FALLBACK_YAML: /home/bp1/executorch/kernels/portable/functions.yaml
-- OUTPUT_DIR: /home/bp1/executorch/cmake-out/configurations
-- Generating selected operator lib:
-- LIB_NAME: optimized_native_cpu_ops_lib
-- OPS_SCHEMA_YAML: /home/bp1/executorch/cmake-out/configurations/merged.yaml
-- ROOT_OPS:
-- INCLUDE_ALL_OPS:
-- OPS_FROM_MODEL:
-- DTYPE_SELECTIVE_BUILD:
Command - /home/bp1/venv/bin/python3;-m;codegen.tools.gen_oplist;--output_path=/home/bp1/executorch/cmake-out/configurations/optimized_native_cpu_ops_lib/selected_operators.yaml;--ops_schema_yaml_path="/home/bp1/executorch/cmake-out/configurations/merged.yaml"
-- Generating kernel bindings:
-- LIB_NAME: optimized_native_cpu_ops_lib
-- FUNCTIONS_YAML: /home/bp1/executorch/cmake-out/configurations/merged.yaml
-- CUSTOM_OPS_YAML:
-- ADD_EXCEPTION_BOUNDARY: TRUE
-- DTYPE_SELECTIVE_BUILD:
Generated files /home/bp1/executorch/cmake-out/configurations/optimized_native_cpu_ops_lib/RegisterCodegenUnboxedKernelsEverything.cpp;/home/bp1/executorch/cmake-out/configurations/optimized_native_cpu_ops_lib/Functions.h;/home/bp1/executorch/cmake-out/configurations/optimized_native_cpu_ops_lib/NativeFunctions.h
-- Generating operator lib:
-- LIB_NAME: optimized_native_cpu_ops_lib
-- KERNEL_LIBS: optimized_kernels;optimized_portable_kernels
-- DEPS: executorch_core
-- DTYPE_SELECTIVE_BUILD:
-- Building for XNNPACK_TARGET_PROCESSOR: arm64
-- Enabling KleidiAI for Arm64
CMake Warning at backends/xnnpack/third-party/XNNPACK/CMakeLists.txt:402 (MESSAGE):
Security: XNNPACK doesn't support BTI control flow integrity with GCC.
Continuing.

-- Generating microkernels.cmake
CMake Warning at extension/llm/tokenizers/third-party/abseil-cpp/CMakeLists.txt:177 (message):
The default and system-level install directories are unsupported except in LTS releases of Abseil. Please set CMAKE_INSTALL_PREFIX to install Abseil in your source or build tree directly.

CMake Deprecation Warning at extension/llm/tokenizers/third-party/sentencepiece/CMakeLists.txt:15 (cmake_minimum_required):
Compatibility with CMake < 3.10 will be removed from a future version of
CMake.

Update the VERSION argument value. Or, use the ... syntax
to tell CMake that the project requires at least but has been updated
to work with policies introduced by or earlier.

-- VERSION: 0.2.1
-- Not Found TCMalloc: TCMALLOC_LIB-NOTFOUND
-- Found atomic: /usr/lib/aarch64-linux-gnu/libatomic.so.1
-- Could NOT find Editline (missing: EDITLINE_INCLUDE_DIR EDITLINE_LIBRARY)

--
-- PCRE2-10.46 configuration summary:

-- Install prefix .................... : /usr/local
-- C compiler ........................ : /usr/bin/cc
-- C compiler flags .................. : -O3 -DNDEBUG

-- Build type ........................ : Release
-- Build 8 bit pcre2 library ......... : ON
-- Build 16 bit pcre2 library ........ : OFF
-- Build 32 bit pcre2 library ........ : OFF
-- Include debugging code ............ : IfDebugBuild
-- Enable JIT compiling support ...... : OFF
-- Use SELinux allocator in JIT ...... : OFF
-- Enable Unicode support ............ : ON
-- Newline char/sequence ............. : LF
-- \R matches only ANYCRLF ........... : OFF
-- \C is disabled .................... : OFF
-- EBCDIC coding ..................... : OFF
-- EBCDIC code for NL ................ : n/a
-- EBCDIC coding ignoring compiler ... : OFF
-- Rebuild char tables ............... : OFF
-- Internal link size ................ : 2
-- Maximum variable lookbehind ....... : 255
-- Nested parentheses limit .......... : 250
-- Heap limit ........................ : 20000000
-- Match limit ....................... : 10000000
-- Match depth limit ................. : MATCH_LIMIT
-- Build shared libs ................. : OFF
-- with symbol versioning ........ : ON
-- Build static libs ................. : ON
-- with PIC enabled .............. : ON
-- Build pcre2grep ................... : OFF
-- Enable JIT in pcre2grep ........... : ON
-- Enable callouts in pcre2grep ...... : ON
-- Enable callout fork in pcre2grep .. : ON
-- Initial buffer size for pcre2grep . : 20480
-- Maximum buffer size for pcre2grep . : 1048576
-- Build tests (implies pcre2test .... : OFF
-- and pcre2grep)
-- Link pcre2grep with libz .......... : ON
-- Link pcre2grep with libbz2 ........ : ON
-- Link pcre2test with libeditline ... : Library not found
-- Link pcre2test with libreadline ... : ON
-- Enable Valgrind support ........... : OFF
-- Use %zu and %td ................... : AUTO

CMake Warning (dev) at extension/llm/tokenizers/CMakeLists.txt:204 (install):
Target re2 has PUBLIC_HEADER files but no PUBLIC_HEADER DESTINATION.
This warning is for project developers. Use -Wno-dev to suppress it.

'/home/bp1/venv/bin/python3' '-c' 'import importlib.util; print(importlib.util.find_spec('torch').submodule_search_locations[0])'
-- Generating selected operator lib:
-- LIB_NAME: quantized_ops_lib
-- OPS_SCHEMA_YAML: /home/bp1/executorch/kernels/quantized/quantized.yaml
-- ROOT_OPS:
-- INCLUDE_ALL_OPS:
-- OPS_FROM_MODEL:
-- DTYPE_SELECTIVE_BUILD:
Command - /home/bp1/venv/bin/python3;-m;codegen.tools.gen_oplist;--output_path=/home/bp1/executorch/cmake-out/kernels/quantized/quantized_ops_lib/selected_operators.yaml;--ops_schema_yaml_path="/home/bp1/executorch/kernels/quantized/quantized.yaml"
-- Generating kernel bindings:
-- LIB_NAME: quantized_ops_lib
-- FUNCTIONS_YAML:
-- CUSTOM_OPS_YAML: /home/bp1/executorch/kernels/quantized/quantized.yaml
-- ADD_EXCEPTION_BOUNDARY: FALSE
-- DTYPE_SELECTIVE_BUILD:
Generated files /home/bp1/executorch/cmake-out/kernels/quantized/quantized_ops_lib/RegisterCodegenUnboxedKernelsEverything.cpp;/home/bp1/executorch/cmake-out/kernels/quantized/quantized_ops_lib/Functions.h;/home/bp1/executorch/cmake-out/kernels/quantized/quantized_ops_lib/NativeFunctions.h;/home/bp1/executorch/cmake-out/kernels/quantized/quantized_ops_lib/RegisterCPUCustomOps.cpp;/home/bp1/executorch/cmake-out/kernels/quantized/quantized_ops_lib/RegisterSchema.cpp;/home/bp1/executorch/cmake-out/kernels/quantized/quantized_ops_lib/CustomOpsNativeFunctions.h
-- Generating operator lib:
-- LIB_NAME: quantized_ops_lib
-- KERNEL_LIBS: quantized_kernels
-- DEPS: executorch_core
-- DTYPE_SELECTIVE_BUILD:
-- --- Configured Options ---

-- BUILD_TESTING : OFF
-- CCACHE_PROGRAM : CCACHE_PROGRAM-NOTFOUND
-- CMAKE_BUILD_TYPE : Release
-- CMAKE_CXX_COMPILER_ID : GNU
-- CMAKE_CXX_STANDARD : 17
-- CMAKE_SYSTEM_PROCESSOR : aarch64
-- CMAKE_TOOLCHAIN_FILE x (unset)
-- EXECUTORCH_BUILD_ARM_BAREMETAL : OFF
-- EXECUTORCH_BUILD_CADENCE : OFF
-- EXECUTORCH_BUILD_COREML : OFF
-- EXECUTORCH_BUILD_CORTEX_M : OFF
-- EXECUTORCH_BUILD_CPUINFO : ON
-- EXECUTORCH_BUILD_DEVTOOLS : OFF
-- EXECUTORCH_BUILD_EXECUTOR_RUNNER : ON
-- EXECUTORCH_BUILD_EXTENSION_APPLE : OFF
-- EXECUTORCH_BUILD_EXTENSION_DATA_LOADER : ON
-- EXECUTORCH_BUILD_EXTENSION_EVALUE_UTIL : ON
-- EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR : ON
-- EXECUTORCH_BUILD_EXTENSION_LLM : ON
-- EXECUTORCH_BUILD_EXTENSION_LLM_APPLE : OFF
-- EXECUTORCH_BUILD_EXTENSION_LLM_RUNNER : ON
-- EXECUTORCH_BUILD_EXTENSION_MODULE : ON
-- EXECUTORCH_BUILD_EXTENSION_NAMED_DATA_MAP : ON
-- EXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL : ON
-- EXECUTORCH_BUILD_EXTENSION_TENSOR : ON
-- EXECUTORCH_BUILD_EXTENSION_TRAINING : OFF
-- EXECUTORCH_BUILD_KERNELS_LLM : ON
-- EXECUTORCH_BUILD_KERNELS_LLM_AOT : OFF
-- EXECUTORCH_BUILD_KERNELS_OPTIMIZED : ON
-- EXECUTORCH_BUILD_KERNELS_QUANTIZED : ON
-- EXECUTORCH_BUILD_KERNELS_QUANTIZED_AOT : OFF
-- EXECUTORCH_BUILD_MPS : OFF
-- EXECUTORCH_BUILD_NEURON : OFF
-- EXECUTORCH_BUILD_OPENVINO : OFF
-- EXECUTORCH_BUILD_PORTABLE_OPS : ON
-- EXECUTORCH_BUILD_PRESET_FILE : /home/bp1/executorch/tools/cmake/preset/linux.cmake
-- EXECUTORCH_BUILD_PTHREADPOOL : ON
-- EXECUTORCH_BUILD_PYBIND : OFF
-- EXECUTORCH_BUILD_QNN : OFF
-- EXECUTORCH_BUILD_SIZE_TEST : OFF
-- EXECUTORCH_BUILD_TESTS : OFF
-- EXECUTORCH_BUILD_TOKENIZERS_WASM : OFF
-- EXECUTORCH_BUILD_VGF : OFF
-- EXECUTORCH_BUILD_VULKAN : OFF
-- EXECUTORCH_BUILD_WASM : OFF
-- EXECUTORCH_BUILD_XNNPACK : ON
-- EXECUTORCH_CADENCE_CPU_RUNNER : OFF
-- EXECUTORCH_COREML_BUILD_EXECUTOR_RUNNER : OFF
-- EXECUTORCH_ENABLE_DTYPE_SELECTIVE_BUILD : FALSE
-- EXECUTORCH_ENABLE_EVENT_TRACER : OFF
-- EXECUTORCH_ENABLE_LOGGING : OFF
-- EXECUTORCH_ENABLE_PROGRAM_VERIFICATION : OFF
-- EXECUTORCH_FLATBUFFERS_MAX_ALIGNMENT : 1024
-- EXECUTORCH_LOG_LEVEL : Info
-- EXECUTORCH_NNLIB_OPT : OFF
-- EXECUTORCH_OPTIMIZE_SIZE : OFF
-- EXECUTORCH_PAL_DEFAULT : posix
-- EXECUTORCH_PAL_DEFAULT_FILE_PATH : /home/bp1/executorch/runtime/platform/default/posix.cpp
-- EXECUTORCH_SELECT_OPS_LIST :
-- EXECUTORCH_SELECT_OPS_MODEL :
-- EXECUTORCH_SELECT_OPS_YAML :
-- EXECUTORCH_USE_CPP_CODE_COVERAGE : OFF
-- EXECUTORCH_USE_DL : ON
-- EXECUTORCH_XNNPACK_ENABLE_KLEIDI : ON
-- EXECUTORCH_XNNPACK_ENABLE_WEIGHT_CACHE : OFF
-- EXECUTORCH_XNNPACK_SHARED_WORKSPACE : ON
-- PYTHON_EXECUTABLE : /home/bp1/venv/bin/python3

---

-- Configuring done (6.6s)
-- Generating done (0.9s)
-- Build files have been written to: /home/bp1/executorch/cmake-out

[cccclai]

I try to understand https://github.com/berndporr/mobilenet_libtorch/tree/main/experimental a bit more, for the loss part, was it the first step you did in the experiment, meaning that you're using pytorch to train? I am not sure executorch position in this flow yet, can you elaborate?

[berndporr]

Sure! The amazing thing with mobilenetv2 is that it consists of a feature detector which in turn consists of a chain of inverted residuals. Then the actual classifier is just a single layer. Turns out that the trained features are pretty much universal. This comes handy in transfer learning where only the classifier layer needs to be re-trained. The parameters of the feature layers can stay frozen. I need to do fast transfer learning on an embedded device. Clearly 90% of all the compute happens in the feature layers which is complex and takes up a lot of CPU time. Executorch is ideal as it's a lot faster than libtorch and has optimised integer operations. It cannot learn but for the classifier I do that in libtorch.

• Feature layers: quatised, fixed parameters in executorch
• Classifier layer: floating point, parameters learn via gradient descent in libtorch

The "experimental" dir does it with the executorch / libtorch combo with quantised feature detectors. There is also a pure libtorch version in the "transfer_learning" dir using floating point. That works fine on both Intel and ARM.