Add example for Espressif ESP32 executorch runner with no optimizations by jpiat · Pull Request #18224 · pytorch/executorch

jpiat · 2026-03-17T07:59:22Z

Summary

This PR introduce a new example for the ESP32 Espressif SoC. The example implement an executorch runner for the ESP32 platform and a project that executes a simple network. The example does not use ops optimized for ESP32 platform but demonstrate feasibility.

Test plan

This example was tested on a ESP32-S3 development platform. The project compiles and when loaded on the platform show the expected log trace.

cc @psiddh @AdrianLundell @digantdesai

…e it included in runner

…play correct error messages

pytorch-bot · 2026-03-17T07:59:27Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18224

📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

Rolling out OSDC (ARC) runners on pull & trunk workflows in PyTorch main

⚠️ 11 Awaiting Approval

As of commit f0db59a with merge base 32a6cec ():

AWAITING APPROVAL - The following workflows need approval before CI can run:

This comment was automatically generated by Dr. CI and updates every 15 minutes.

meta-cla · 2026-03-17T07:59:28Z

Hi @jpiat!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks!

Copilot

Pull request overview

Adds an Espressif (ESP32 / ESP32-S3) ExecuTorch “executor runner” example project plus a new CMake preset to help cross-compile ExecuTorch for ESP targets.

Changes:

Introduces a new esp-baremetal CMake preset and preset file for ESP-oriented ExecuTorch builds.
Adds a full ESP-IDF example project (runner component + template project + build script + docs).
Adds ESP-IDF configuration defaults (sdkconfig defaults + partition table) for ESP32-S3-oriented setups.

Reviewed changes

Copilot reviewed 17 out of 17 changed files in this pull request and generated 11 comments.

Show a summary per file

File	Description
tools/cmake/preset/esp_baremetal.cmake	New ESP-oriented ExecuTorch build preset settings.
CMakePresets.json	Registers the `esp-baremetal` configure preset.
examples/espressif/README.md	End-to-end instructions for exporting, building, flashing, and troubleshooting on ESP32/ESP32-S3.
examples/espressif/build.sh	Helper script to build the ESP-IDF project and optionally generate `model_pte.h`.
examples/espressif/executor_runner/CMakeLists.txt	ESP-IDF component + standalone build definition for the runner.
examples/espressif/executor_runner/esp_executor_runner.cpp	Main ESP runner implementation (PAL hooks, loading, execution, optional FS load/ETDump/BundleIO).
examples/espressif/executor_runner/esp_memory_allocator.{h,cpp}	ESP-specific allocator wrapper with usage tracking.
examples/espressif/executor_runner/esp_perf_monitor.{h,cpp}	Simple perf measurements using cycle counter + `esp_timer`.
examples/espressif/executor_runner/pte_to_header.py	Converts `.pte` into a C header for compiled-in model embedding.
examples/espressif/project/CMakeLists.txt	Template ESP-IDF project that pulls in the runner component.
examples/espressif/project/main/{CMakeLists.txt,main.cpp}	Minimal “main” component that calls into the runner.
examples/espressif/project/sdkconfig.defaults	Default ESP-IDF configuration for the example (generic).
examples/espressif/project/sdkconfig.defaults.esp32s3	Target-specific ESP32-S3 config defaults.
examples/espressif/project/partitions.csv	Custom partition table for the example firmware.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

jpiat · 2026-03-17T08:17:48Z

@pytorchbot label "release notes: examples"

meta-cla · 2026-03-17T08:17:57Z

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Adds a new Espressif ESP32/ESP32-S3 example integration for running ExecuTorch models under ESP-IDF, along with build tooling and documentation to cross-compile ExecuTorch for Xtensa and build/flash an example project.

Changes:

Introduces an ESP-IDF executor_runner component and example project scaffolding (CMake, app_main, sdkconfig defaults, partitions).
Adds an ESP-focused CMake preset (esp-baremetal) to configure ExecuTorch builds for the target.
Adds a model conversion helper (pte_to_header.py), a build helper script, and a comprehensive README.

Reviewed changes

Copilot reviewed 17 out of 17 changed files in this pull request and generated 14 comments.

Show a summary per file

File	Description
tools/cmake/preset/esp_baremetal.cmake	New preset to configure ExecuTorch build options for ESP targets.
CMakePresets.json	Registers the `esp-baremetal` CMake preset.
examples/espressif/README.md	End-to-end documentation for exporting, building, flashing, and troubleshooting.
examples/espressif/build.sh	Helper script to generate `model_pte.h` and build via `idf.py`.
examples/espressif/executor_runner/CMakeLists.txt	ESP-IDF component definition + standalone CMake build option; integrates prebuilt ExecuTorch libs.
examples/espressif/executor_runner/esp_executor_runner.cpp	Core runner implementation: loading PTE, setting up allocators, executing, logging, optional SPIFFS load.
examples/espressif/executor_runner/esp_memory_allocator.{h,cpp}	Custom allocator wrapper with tracking helpers.
examples/espressif/executor_runner/esp_perf_monitor.{h,cpp}	Perf measurement helpers using cycle counter and `esp_timer`.
examples/espressif/executor_runner/pte_to_header.py	Converts a `.pte` into a C header byte array for firmware embedding.
examples/espressif/project/CMakeLists.txt	Example ESP-IDF project top-level CMake.
examples/espressif/project/main/CMakeLists.txt	Minimal `main` component that depends on `executor_runner`.
examples/espressif/project/main/main.cpp	Provides `app_main()` that invokes `executor_runner_main()`.
examples/espressif/project/sdkconfig.defaults	Baseline ESP-IDF defaults for the example project.
examples/espressif/project/sdkconfig.defaults.esp32s3	ESP32-S3-specific defaults (PSRAM, etc.).
examples/espressif/project/partitions.csv	Custom partition table template for the example project.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.

… on xtensa.

…argument.

Copilot

Pull request overview

Adds a new Espressif ESP32/ESP32-S3 example that demonstrates running an ExecuTorch .pte on ESP-IDF (no ESP32-optimized ops), plus a CMake preset to cross-compile ExecuTorch for the target.

Changes:

Introduces an ESP-IDF-based executor runner component + template project (runner API, PAL, perf monitor, memory allocator, build glue).
Adds an Espressif example README and helper script for building/flashing and converting .pte → model_pte.h.
Adds a new esp-baremetal CMake preset and a small threadpool type fix.

Reviewed changes

Copilot reviewed 20 out of 20 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
tools/cmake/preset/esp_baremetal.cmake	New preset defaults for building ExecuTorch for ESP cross-compilation.
CMakePresets.json	Registers the new `esp-baremetal` preset.
extension/threadpool/threadpool.cpp	Adjusts tsan thread limit type to match `result` type.
examples/espressif/README.md	End-to-end documentation for exporting, building, flashing, and troubleshooting.
examples/espressif/build.sh	Helper script to generate model header and build the ESP-IDF project.
examples/espressif/project/CMakeLists.txt	ESP-IDF project wrapper that pulls in the runner component.
examples/espressif/project/main/CMakeLists.txt	Minimal main component depending on the runner.
examples/espressif/project/main/main.cpp	ESP-IDF `app_main()` entrypoint that invokes the runner.
examples/espressif/project/sdkconfig.defaults	Baseline ESP-IDF defaults (ESP32/ESP32-S3).
examples/espressif/project/sdkconfig.defaults.esp32s3	ESP32-S3-specific sdkconfig defaults (Octal PSRAM, etc.).
examples/espressif/project/partitions.csv	Example partition table for the template project.
examples/espressif/executor_runner/CMakeLists.txt	ESP-IDF component definition + integration with prebuilt ExecuTorch libs + optional `.pte` header generation.
examples/espressif/executor_runner/esp_executor_runner.h	Public C API for initializing, setting inputs, executing, and reading outputs.
examples/espressif/executor_runner/esp_executor_runner.cpp	Core runner implementation: loading PTE, allocators, inputs/outputs, optional BundleIO/ETDump, optional filesystem load.
examples/espressif/executor_runner/esp_pal.cpp	ESP PAL overrides (ticks, log emission, init/abort).
examples/espressif/executor_runner/esp_memory_allocator.h	Wrapper allocator tracking used/free bytes for instrumentation.
examples/espressif/executor_runner/esp_memory_allocator.cpp	Implementation of the tracking allocator.
examples/espressif/executor_runner/esp_perf_monitor.h	Perf monitor API for ESP.
examples/espressif/executor_runner/esp_perf_monitor.cpp	Cycle/time based measurement and reporting for ESP.
examples/espressif/executor_runner/pte_to_header.py	Utility to embed `.pte` into firmware as a C byte array header.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-03-23T12:34:28Z

+  if (!ctx.method->ok()) {
+    ET_LOG(
+        Info,
+        "Loading of method %s failed with status 0x%" PRIx32,
+        ctx.method_name,
+        static_cast<unsigned long>(ctx.method->error()));
+  }


runner_init() logs when program.load_method() fails but then continues and dereferences *ctx.method.value() (e.g., in input preparation). If the Result is not ok, this can trigger an assert/abort or undefined behavior. Return early (or ET_CHECK_MSG(ctx.method->ok(), ...)) before using the Method when load fails.

Suggested change

if (!ctx.method->ok()) {

ET_LOG(

Info,

"Loading of method %s failed with status 0x%" PRIx32,

ctx.method_name,

static_cast<unsigned long>(ctx.method->error()));

}

ET_CHECK_MSG(

ctx.method->ok(),

"Loading of method %s failed with status 0x%" PRIx32,

ctx.method_name,

static_cast<unsigned long>(ctx.method->error()));

Copilot · 2026-03-23T12:34:29Z

+    ET_LOG(Fatal, "Failed to load model from filesystem.");
+    return false;
+  }
+  model_pte = buffer;


When FILESYSTEM_LOAD is enabled, the model bytes are allocated from method_allocation_pool via file_allocator, but the same method_allocation_pool is then reused for ctx.method_allocator allocations in runner_init(). This can overwrite the PTE buffer that BufferDataLoader/Program still references and corrupt execution. Load the PTE into separate storage (heap/PSRAM), or reserve an initial region in the pool and start the method allocator after it, or switch to a file-backed DataLoader.

Suggested change

model_pte = buffer;

// Copy the loaded PTE into separate heap/PSRAM storage to avoid it being

// overwritten when method_allocation_pool is reused by runner_init().

void* heap_pte = std::malloc(buffer_size);

if (heap_pte == nullptr) {

ET_LOG(Fatal, "Failed to allocate heap memory for model PTE.");

return false;

}

std::memcpy(heap_pte, buffer, buffer_size);

model_pte = static_cast<uint8_t*>(heap_pte);

Copilot · 2026-03-23T12:34:29Z

+        static_cast<unsigned long>(buffer_bytes),
+        static_cast<unsigned long>(tensor_bytes));
+    return false;
+  }


et_runner_get_output() validates buffer_bytes but never checks that buffer is non-null before calling memcpy(). If callers pass buffer=nullptr with a non-zero size, this will crash. Add an explicit null check (e.g., require non-null when tensor_bytes > 0).

Suggested change

}

}

if (tensor_bytes > 0 && buffer == nullptr) {

ET_LOG(

Error,

"Output %lu: null buffer passed for non-empty tensor (size=%lu bytes).",

static_cast<unsigned long>(output_idx),

static_cast<unsigned long>(tensor_bytes));

return false;

}

Copilot · 2026-03-23T12:34:30Z

+| Chip     | CPU           | Internal SRAM | PSRAM (optional) |
+|----------|---------------|---------------|------------------|
+| ESP32    | Xtensa LX6 (dual-core, 240MHz) | ~520KB | 4-8MB |
+| ESP32-S3 | Xtensa LX7 (dual-core, 240MHz) | ~512KB | 2-32MB (Octal) |
+


The markdown table under “Supported Targets” has an extra leading | on each row (|| ...), which renders as an empty first column in GitHub markdown. Use standard table syntax with a single leading pipe per row.

Copilot · 2026-03-23T12:34:30Z

+# Name, Type, SubType, Offset, Size, Flags
+nvs,data,nvs,0x9000,24K,
+phy_init,data,phy,0xf000,4K,
+factory,app,factory,0x10000,2M,


The example code enables SPIFFS-based model loading (partition label storage), and the README suggests adding a SPIFFS partition, but the provided partitions.csv doesn’t include a storage,data,spiffs,... entry. Consider adding a commented-out SPIFFS row (or include it by default) so users enabling FILESYSTEM_LOAD don’t hit SPIFFS partition not found.

Suggested change

factory,app,factory,0x10000,2M,

factory,app,factory,0x10000,2M,

# Optional SPIFFS partition for FILESYSTEM_LOAD (label must be "storage")

# storage,data,spiffs,,1M,

Copilot · 2026-03-23T12:34:30Z

+set_overridable_option(EXECUTORCH_BUILD_KERNELS_QUANTIZED ON)
+set_overridable_option(EXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL ON)
+set_overridable_option(EXECUTORCH_ENABLE_LOGGING ON)
+set_overridable_option(EXECUTORCH_BUILD_ARM_ETDUMP OFF)


This preset uses set_overridable_option(EXECUTORCH_BUILD_ARM_ETDUMP OFF), which stores the value as a CACHE STRING with no description. In tools/cmake/preset/arm_baremetal.cmake, the same knob is defined via define_overridable_option(... BOOL ...), which gives it the right type and makes it discoverable. Consider switching to define_overridable_option here as well (or use a non-Arm-specific option name).

Suggested change

set_overridable_option(EXECUTORCH_BUILD_ARM_ETDUMP OFF)

define_overridable_option(

EXECUTORCH_BUILD_ARM_ETDUMP

OFF

BOOL

"Enable ETDump devtools support for bare-metal targets"

)

nil-is-all · 2026-04-01T17:32:59Z

@jpiat could you resolve the lintrunner errors reported?

Copilot

Pull request overview

Copilot reviewed 20 out of 20 changed files in this pull request and generated 4 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-02T08:51:55Z

+  EspMemoryAllocator file_allocator(
+      method_allocation_pool_size, method_allocation_pool);
+  auto [buffer, buffer_size] =
+      load_file_from_fs("/spiffs/model.pte", file_allocator);


With FILESYSTEM_LOAD enabled, file_allocator loads /spiffs/model.pte into method_allocation_pool, and model_pte points into that buffer. runner_init() then constructs ctx.method_allocator over the same method_allocation_pool starting at offset 0, so subsequent allocations can overwrite the model bytes and corrupt Program::load()/method execution. Fix by ensuring the model storage is not reused by the method allocator (e.g., partition the pool and start the method allocator after the aligned model size, or allocate the model in a separate heap/PSRAM buffer, or switch to a file-backed DataLoader).

This remains a problem, as the method allocator could wind up overwriting the PTE file when it creates a new MemoryAllocator for the method_allocation_pool. I think there are two options:

Split the pool in two pools: one for the PTE file, one for the method allocator

Share the same MemoryAllocator object for both uses, so that when method allocation starts, it uses memory after the PTE file, instead of overwriting.

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 20 out of 20 changed files in this pull request and generated 3 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-02T11:50:38Z

+        Info,
+        "Loading of method %s failed with status 0x%" PRIx32,
+        ctx.method_name,
+        static_cast<unsigned long>(ctx.method->error()));


This log uses a 0x%" PRIx32 format specifier but passes static_cast<unsigned long>(ctx.method->error()). Because ET_LOG is variadic, this type mismatch is undefined behavior on some ABIs and can print garbage. Cast the argument to uint32_t (or change the format to match the argument type).

Suggested change

static_cast<unsigned long>(ctx.method->error()));

static_cast<uint32_t>(ctx.method->error()));

Copilot · 2026-04-02T11:50:39Z

+      status == Error::Ok,
+      "Execution of method %s failed with status 0x%" PRIx32,
+      ctx.method_name,
+      static_cast<unsigned long>(status));


ET_CHECK_MSG uses a 0x%" PRIx32 format specifier but passes static_cast<unsigned long>(status). This is a variadic type mismatch and can lead to incorrect output/UB. Cast to uint32_t (or update the format specifier) consistently with other PRIx32 usages in this file.

Suggested change

static_cast<unsigned long>(status));

static_cast<uint32_t>(status));

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 20 out of 20 changed files in this pull request and generated 5 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-02T15:48:31Z

+  EspMemoryAllocator file_allocator(
+      method_allocation_pool_size, method_allocation_pool);
+  auto [buffer, buffer_size] =
+      load_file_from_fs("/spiffs/model.pte", file_allocator);
+  if (buffer == nullptr) {
+    ET_LOG(Fatal, "Failed to load model from filesystem.");
+    return false;
+  }
+  model_pte = buffer;
+  model_pte_size = buffer_size;
+  pte_size = buffer_size;
+#else
+  pte_size = sizeof(model_pte);
+#endif


When FILESYSTEM_LOAD is enabled, the model is loaded into method_allocation_pool via file_allocator, but then runner_init() constructs ctx.method_allocator over the same method_allocation_pool starting at offset 0. Subsequent allocations will overwrite model_pte and can corrupt Program::load() / method metadata that references the buffer. Reserve the model region in the pool (e.g., construct EspMemoryAllocator for runtime allocations with base_address = method_allocation_pool + aligned_model_size and size = pool_size - aligned_model_size), or load into separate storage (heap/PSRAM) that won’t be reused by the method allocator.

Copilot · 2026-04-02T15:48:31Z

+void* et_pal_allocate(ET_UNUSED size_t size) {
+  return nullptr;
+}
+
+void et_pal_free(ET_UNUSED void* ptr) {}


et_pal_allocate()/et_pal_free() are overridden to always return nullptr / no-op. This disables ExecuTorch’s PAL-backed allocations even though this example uses heap-backed features elsewhere (e.g., std::vector, filesystem I/O). Implement these using malloc/free (or heap_caps_malloc/heap_caps_free with appropriate caps) so runtime components that rely on PAL allocation (e.g., PlatformMemoryAllocator or temp allocations when no temp allocator is provided) work correctly.

Suggested change

void* et_pal_allocate(ET_UNUSED size_t size) {

return nullptr;

}

void et_pal_free(ET_UNUSED void* ptr) {}

void* et_pal_allocate(size_t size) {

#if defined(ESP_PLATFORM)

// Allocate from the default-capability heap on ESP platforms.

return heap_caps_malloc(size, MALLOC_CAP_DEFAULT);

#else

// Fallback to standard malloc on non-ESP platforms.

return malloc(size);

#endif

}

void et_pal_free(void* ptr) {

#if defined(ESP_PLATFORM)

heap_caps_free(ptr);

#else

free(ptr);

#endif

}

Copilot · 2026-04-02T15:48:32Z

+| Chip     | CPU           | Internal SRAM | PSRAM (optional) |
+|----------|---------------|---------------|------------------|
+| ESP32    | Xtensa LX6 (dual-core, 240MHz) | ~520KB | 4-8MB |
+| ESP32-S3 | Xtensa LX7 (dual-core, 240MHz) | ~512KB | 2-32MB (Octal) |
+


The Markdown tables use a double leading pipe (|| ... |), which creates an empty first column in GitHub’s renderer. Use a single leading pipe for each row (e.g., | Chip | CPU | ... |) and update the separator row accordingly so the table renders correctly.

Copilot · 2026-04-02T15:48:32Z

+extern void executor_runner_main(void);
+
+extern "C" void app_main(void) {
+  printf("Starting executorch runner !\n");


Log message has inconsistent capitalization/spacing ("executorch" and a space before !). Consider changing to a consistent product name and punctuation (e.g., "Starting ExecuTorch runner!\n") to match the rest of the example output in the README.

Suggested change

printf("Starting executorch runner !\n");

printf("Starting ExecuTorch runner!\n");

Copilot · 2026-04-02T15:48:32Z

+  idf_component_register(
+    SRCS
+    "esp_executor_runner.cpp"
+    "esp_pal.cpp"
+    "esp_memory_allocator.cpp"
+    "esp_perf_monitor.cpp"
+    INCLUDE_DIRS
+    "."
+    REQUIRES
+    esp_timer
+    esp_system
+    spiffs
+  )


idf_component_register() lists spiffs in REQUIRES unconditionally, but the code only includes/uses SPIFFS when FILESYSTEM_LOAD is enabled. Consider making the spiffs dependency conditional on FILESYSTEM_LOAD so the default (compiled-in model) build doesn’t pull in an unnecessary component or require SPIFFS-related sdkconfig options.

jpiat · 2026-04-03T14:41:31Z

@jpiat could you resolve the lintrunner errors reported?

should be fixed now

Copilot

Pull request overview

Copilot reviewed 20 out of 20 changed files in this pull request and generated 5 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

+  EspMemoryAllocator file_allocator(
+      method_allocation_pool_size, method_allocation_pool);
+  auto [buffer, buffer_size] =
+      load_file_from_fs("/spiffs/model.pte", file_allocator);
+  if (buffer == nullptr) {
+    ET_LOG(Fatal, "Failed to load model from filesystem.");
+    return false;
+  }
+  model_pte = buffer;
+  model_pte_size = buffer_size;
+  pte_size = buffer_size;


+  for (int i = 0; i < outputs.size(); ++i) {
+    if (outputs[i].isTensor()) {
+      Tensor tensor = outputs[i].toTensor();
+#if defined(ET_LOG_DUMP_OUTPUT)
+      for (int j = 0; j < tensor.numel(); ++j) {
+        if (tensor.scalar_type() == ScalarType::Int) {
+          printf(
+              "Output[%d][%d]: (int) %d\n",
+              i,
+              j,
+              tensor.const_data_ptr<int>()[j]);
+        } else if (tensor.scalar_type() == ScalarType::Float) {
+          printf(
+              "Output[%d][%d]: (float) %f\n",
+              i,
+              j,
+              tensor.const_data_ptr<float>()[j]);
+        } else if (tensor.scalar_type() == ScalarType::Char) {
+          printf(
+              "Output[%d][%d]: (char) %d\n",
+              i,
+              j,
+              tensor.const_data_ptr<int8_t>()[j]);
+        } else if (tensor.scalar_type() == ScalarType::Bool) {
+          printf(
+              "Output[%d][%d]: (bool) %s (0x%x)\n",


+    for (int i = 0; i < inputs.size(); ++i) {
+      if (inputs[i].isTensor()) {
+        Tensor tensor = inputs[i].toTensor();
+        for (int j = 0; j < tensor.numel(); ++j) {
+          if (tensor.scalar_type() == ScalarType::Int) {


+      ET_LOG(
+          Error,
+          "ETDump: Could not set_debug_buffer() error:0x%" PRIx32,
+          result.error());


+    ET_CHECK_MSG(
+        status == Error::Ok,
+        "load_bundled_input failed with status 0x%" PRIx32,
+        status);


jpiat added 9 commits March 17, 2026 08:53

Added an example executor runner for espressif

b21a1fd

Changed build to generate the model header in the project dir and hav…

52f13f3

…e it included in runner

Fixed some compile error. Should probably be fixed differently to dis…

dee4c8d

…play correct error messages

Added default configuration for esp-baremetal

1978162

Fixed README for lib compilation

7d6a0de

Updated CMake to cope with executorch find package issue in esp idf

fcffd84

First inference ran ! Need a ESP32 with at least 8MB of RAM.

1d89f39

Adjusted example for selective build to save on flash

0ec941d

Modified README to inform user about which platform is really supported

0f29967

jpiat requested a review from larryliu0820 as a code owner March 17, 2026 07:59

Copilot AI review requested due to automatic review settings March 17, 2026 07:59

jpiat requested a review from kirklandsign as a code owner March 17, 2026 07:59

Copilot started reviewing on behalf of jpiat March 17, 2026 08:00 View session

Copilot AI reviewed Mar 17, 2026

View reviewed changes

pytorch-bot Bot added the release notes: examples Changes to any of our example LLMs integrations, such as Llama3 and Llava label Mar 17, 2026

meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Mar 17, 2026

Fix README.md for clearer instructions

c363a73

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings March 17, 2026 08:24

Copilot started reviewing on behalf of jpiat March 17, 2026 08:24 View session

Copilot AI reviewed Mar 17, 2026

View reviewed changes

jpiat added 2 commits March 17, 2026 09:38

Applied some of the suggested copilot modifications

1a7b28b

Fixing issue with min function not taking different type on its input…

3a3c888

… on xtensa.

Copilot AI review requested due to automatic review settings March 17, 2026 10:32

jpiat requested a review from kimishpatel as a code owner March 17, 2026 10:32

Copilot started reviewing on behalf of jpiat March 17, 2026 10:32 View session

Casting result of processors_count instead of forcing type on second …

ba1c8d4

…argument.

Copilot AI review requested due to automatic review settings March 23, 2026 12:28

Copilot started reviewing on behalf of jpiat March 23, 2026 12:29 View session

Copilot AI reviewed Mar 23, 2026

View reviewed changes

jpiat added 2 commits April 2, 2026 09:55

Fixed linter errors

195277f

Merge branch 'main' into espressif-example

c68e856

Copilot AI review requested due to automatic review settings April 2, 2026 08:44

Copilot started reviewing on behalf of jpiat April 2, 2026 08:44 View session

Copilot AI reviewed Apr 2, 2026

View reviewed changes

Update tools/cmake/preset/esp_baremetal.cmake

c10a25e

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings April 2, 2026 11:44

jpiat and others added 2 commits April 2, 2026 13:44

Update examples/espressif/README.md

054ad99

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Update examples/espressif/executor_runner/pte_to_header.py

ee8ee17

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot started reviewing on behalf of jpiat April 2, 2026 11:45 View session

Copilot AI reviewed Apr 2, 2026

View reviewed changes

Update examples/espressif/executor_runner/esp_pal.cpp

197ea6a

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings April 2, 2026 15:42

Copilot started reviewing on behalf of jpiat April 2, 2026 15:43 View session

Copilot AI reviewed Apr 2, 2026

View reviewed changes

Fixed compile error with overridable not properly defined

ad2db6d

Fixed linter issue on non python files

09a9f9e

Copilot AI review requested due to automatic review settings April 14, 2026 13:26

Copilot started reviewing on behalf of jpiat April 14, 2026 13:27 View session

Copilot AI reviewed Apr 14, 2026

View reviewed changes

Merge branch 'main' into espressif-example

83f02b8

nil-is-all assigned rascani Apr 22, 2026

jpiat added 3 commits April 27, 2026 11:06

Removed reference to non existing LICENSE file

1fad16e

Merge branch 'main' into espressif-example

6ab1d1f

Removed newline at end of file

f0db59a

-  model_pte = buffer;
+  // Copy the loaded PTE into separate heap/PSRAM storage to avoid it being
+  // overwritten when method_allocation_pool is reused by runner_init().
+  void* heap_pte = std::malloc(buffer_size);
+  if (heap_pte == nullptr) {
+    ET_LOG(Fatal, "Failed to allocate heap memory for model PTE.");
+    return false;
+  }
+  std::memcpy(heap_pte, buffer, buffer_size);
+  model_pte = static_cast<uint8_t*>(heap_pte);

-  }
+  }
+  if (tensor_bytes > 0 && buffer == nullptr) {
+    ET_LOG(
+        Error,
+        "Output %lu: null buffer passed for non-empty tensor (size=%lu bytes).",
+        static_cast<unsigned long>(output_idx),
+        static_cast<unsigned long>(tensor_bytes));
+    return false;
+  }

	static_cast<unsigned long>(ctx.method->error()));
	static_cast<uint32_t>(ctx.method->error()));

	static_cast<unsigned long>(status));
	static_cast<uint32_t>(status));

-void* et_pal_allocate(ET_UNUSED size_t size) {
-  return nullptr;
-}
-void et_pal_free(ET_UNUSED void* ptr) {}
+void* et_pal_allocate(size_t size) {
+#if defined(ESP_PLATFORM)
+  // Allocate from the default-capability heap on ESP platforms.
+  return heap_caps_malloc(size, MALLOC_CAP_DEFAULT);
+#else
+  // Fallback to standard malloc on non-ESP platforms.
+  return malloc(size);
+#endif
+}
+void et_pal_free(void* ptr) {
+#if defined(ESP_PLATFORM)
+  heap_caps_free(ptr);
+#else
+  free(ptr);
+#endif
+}

	printf("Starting executorch runner !\n");
	printf("Starting ExecuTorch runner!\n");

Conversation

jpiat commented Mar 17, 2026 • edited by pytorch-bot Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Uh oh!

pytorch-bot Bot commented Mar 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18224

❗ 1 Active SEVs

⚠️ 11 Awaiting Approval

Uh oh!

meta-cla Bot commented Mar 17, 2026

Action Required

Process

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jpiat commented Mar 17, 2026

Uh oh!

meta-cla Bot commented Mar 17, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

nil-is-all commented Apr 1, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

jpiat commented Mar 17, 2026 •

edited by pytorch-bot Bot

Loading

pytorch-bot Bot commented Mar 17, 2026 •

edited

Loading