80 Pull requests merged by 23 people

• Qualcomm AI Engine Direct - Ensure that math invariant ops don't change scale and offset

  #11989 merged Jun 29, 2025

• Correct debug handle merging logic

  #12073 merged Jun 29, 2025

• Arm backend: Add VGF tests to add op unit tests

  #12058 merged Jun 28, 2025

• Skip exp float16 test in XNNPACk

  #12097 merged Jun 28, 2025

• Fix llava test target cmake

  #12096 merged Jun 28, 2025

• Introduce TextLLMRunner - sync updates from fbsource

  #12094 merged Jun 28, 2025

• Unbreak build_size_test.sh on Mac

  #12089 merged Jun 28, 2025

• Fix apple frameworks by patching header

  #12092 merged Jun 28, 2025

• Intorduce TextLLMRunner.

  #12055 merged Jun 28, 2025

• [ET][Portable] Test bad alpha values: op_add

  #12091 merged Jun 28, 2025

• [ET][Portable] Test bad alpha values: op_add

  #12030 merged Jun 28, 2025

• Revert "Milestone2.1: Partition to_dim_order_copy op in XNN delegate"

  #12090 merged Jun 28, 2025

• Fix LLaMA demo app.

  #12079 merged Jun 28, 2025

• [ET][Portable] Check scalar overflow: op_leaky_relu

  #12086 merged Jun 28, 2025

• [ET][Portable] Check scalar overflow: op_hardtanh

  #12085 merged Jun 28, 2025

• [ET][Portable] Check scalar overflow: op_scatter

  #12084 merged Jun 28, 2025

• [ET][Portable] Check scalar overflow: op_constant_pad_nd

  #12083 merged Jun 28, 2025

• [ET][Portable] Check scalar overflow: op_scalar_tensor

  #12082 merged Jun 28, 2025

• [ET][Portable] Check scalar overflow: op_full_like

  #12081 merged Jun 28, 2025

• [ET][Portable] Check scalar overflow: op_full

  #12080 merged Jun 28, 2025

• unbreak lintrunner in backends/arm

  #12075 merged Jun 28, 2025

• Detect release builds by comparing CMAKE_BUILD_TYPE to Debug, not Release

  #12045 merged Jun 28, 2025

• Show output of the size command in size test scripts

  #12018 merged Jun 28, 2025

• [ET][Portable] Check scalar overflow: op_leaky_relu

  #12027 merged Jun 27, 2025

• [ET][Portable] Check scalar overflow: op_hardtanh

  #12029 merged Jun 27, 2025

• [ET][Portable] Check scalar overflow: op_scatter

  #12028 merged Jun 27, 2025

• [ET][Portable] Check scalar overflow: op_constant_pad_nd

  #12026 merged Jun 27, 2025

• [ET][Portable] Check scalar overflow: op_scalar_tensor

  #12025 merged Jun 27, 2025

• [ET][Portable] Check scalar overflow: op_full_like

  #12015 merged Jun 27, 2025

• [ET][Portable] Check scalar overflow: op_full

  #12014 merged Jun 27, 2025

• Sync LLaMARunner.h with Meta internal copy

  #12076 merged Jun 27, 2025

• Bump pinned PT commit to 20250625 and mirror torch/standalone to ExecuTorch

  #12068 merged Jun 27, 2025

• Refactor debug_handles type to use DebugHandles type alias

  #12061 merged Jun 27, 2025

• Unbreak build_size_test.sh (-Wvla-cxx-extension) on my local machine

  #12017 merged Jun 27, 2025

• [ET-VK][ez] Handle zero-element tensors when building Vulkan graph

  #12021 merged Jun 27, 2025

• Make hydra and omegaconf core dependency

  #12060 merged Jun 27, 2025

• Fix jni_layer_llama.cpp generate_from_pos missing return

  #12066 merged Jun 27, 2025

• Bump pinned PT commit to 20250625 and mirror torch/standalone to ExecuTorch

  #11968 merged Jun 27, 2025

• Milestone2.1: Partition to_dim_order_copy op in XNN delegate

  #11286 merged Jun 27, 2025

• Arm backend: Print output to log when using BundleIO

  #12000 merged Jun 27, 2025

• [ET-VK][ez] Explicitly skip marking output nodes that are mutable buffers

  #12020 merged Jun 27, 2025

• [llm] Support different shape of input_pos

  #11966 merged Jun 27, 2025

• Fix memory planning algo for blocked mem IDs.

  #11969 merged Jun 27, 2025

• Arm backend: Fix op_index_tensor's ADD accumulator

  #12056 merged Jun 27, 2025

• [diff-train-patch] apply patch to op_fill_test

  #12057 merged Jun 27, 2025

• Use extension/llm directly from LLaMARunner.

  #12050 merged Jun 27, 2025

• Prim ops replace abort with error code

  #11985 merged Jun 27, 2025

• Remove unnecessary validation for local global attention

  #12049 merged Jun 27, 2025

• expose reinplace pass in lowering

  #12044 merged Jun 27, 2025

• Split LLaMA and LLaVA runners.

  #12036 merged Jun 27, 2025

• [ET][Test] Scalar overflow test macro

  #12043 merged Jun 27, 2025

• [ET][Portable] Check scalar overflow: op_fill

  #12042 merged Jun 27, 2025

• [ET][Portable] Add util to check for overflow when casting scalar

  #12041 merged Jun 27, 2025

• [EE/BE][ET][Portable] Eliminate usage of ET_SWITCH_SCALAR_OBJ_TYPES in portable kernels

  #12040 merged Jun 27, 2025

• [ET][Optimized] Eliminate usage of ET_SWITCH_SCALAR in optimized kernels

  #12046 merged Jun 27, 2025

• [EE/BE][ET][Portable] Move scalar_to utils to scalar_utils.h

  #12035 merged Jun 27, 2025

• [ET][Optimized] Eliminate usage of ET_SWITCH_SCALAR in optimized kernels

  #12033 merged Jun 27, 2025

• Benchmark Gemma-3

  #11978 merged Jun 27, 2025

• Pin bump for optimum-executorch

  #12022 merged Jun 27, 2025

• insert write back now tries to avoid inserting if all operations were inplace

  #12008 merged Jun 27, 2025

• [ET][Test] Scalar overflow test macro

  #12013 merged Jun 27, 2025

• [ET][Portable] Check scalar overflow: op_fill

  #12012 merged Jun 27, 2025

• [ET][Portable] Add util to check for overflow when casting scalar

  #12011 merged Jun 27, 2025

• [EE/BE][ET][Portable] Eliminate usage of ET_SWITCH_SCALAR_OBJ_TYPES in portable kernels

  #12010 merged Jun 27, 2025

• Remove obsolete tests from LLaMaRunner.

  #12034 merged Jun 26, 2025

• Move LlmConfig to extension/llm

  #11872 merged Jun 26, 2025

• Support while in pass base

  #12004 merged Jun 26, 2025

• [EE/BE][ET][Portable] Move scalar_to utils to scalar_utils.h

  #12009 merged Jun 26, 2025

• Bump tokenizers to incorporate new features for HF tokenizer

  #11995 merged Jun 26, 2025

• Update MemoryPlanning Verifier to only assume model has user input if it has at least one tensor input

  #10617 merged Jun 26, 2025

• Add idma Fake operators for AoT.

  #11867 merged Jun 26, 2025

• [llm] Route generate_from_pos() JNI API to native llm runner API

  #11952 merged Jun 26, 2025

• Add instructions to enable ccache

  #11838 merged Jun 26, 2025

• Fix mutable buffer tagging in ExecuTorch

  #11990 merged Jun 26, 2025

• Update XNNPACK to CMake Fix

  #11856 merged Jun 26, 2025

• [xnn update prep] Update ET Pthreadpool third-party to google

  #11507 merged Jun 26, 2025

• [xnn update prep] deprecate sdpa

  #11506 merged Jun 26, 2025

• [ET-VK][ez] Handle zero-element tensors when building Vulkan graph

  #11984 merged Jun 26, 2025

• [ET-VK][ez] Explicitly skip marking output nodes that are mutable buffers

  #11983 merged Jun 26, 2025

• Integrated debug_handle to operator name mapping into inspector

  #12001 merged Jun 26, 2025

21 Pull requests opened by 18 people

• Add Pybindings for Program.h/cpp

  #12016 opened Jun 26, 2025

• Specialize BroadcastIndexesRange for the case where there is only 1 contiguous input

  #12023 opened Jun 26, 2025

• Not depending on examples/models/llama/runner/ APIs

  #12032 opened Jun 26, 2025

• Introduce sym_max and sym_min ops to executorch

  #12037 opened Jun 26, 2025

• Add gtest macros and matchers for Result<T> and Error.

  #12038 opened Jun 26, 2025

• Add llm runner into apple frameworks build

  #12047 opened Jun 27, 2025

• Enable Gemma3 1B on ExecuTorch

  #12048 opened Jun 27, 2025

• Add code to Bundleio to generate error stats

  #12051 opened Jun 27, 2025

• USe -WShadow in clang builds

  #12052 opened Jun 27, 2025

• Qualcomm AI Engine Direct - DRAFT for GA QWEN2.5 0.5B

  #12054 opened Jun 27, 2025

• Fixes in to_executorch for while

  #12062 opened Jun 27, 2025

• Add export recipes for xnnpack (#12069)

  #12070 opened Jun 27, 2025

• fix is_inplace_node check

  #12071 opened Jun 27, 2025

• Pass Dependencies When Proposing Partitions VIA New Group-Based Partitioner

  #12072 opened Jun 27, 2025

• Update test_ring_attention.py to fix CI

  #12074 opened Jun 27, 2025

• Added Command Line Option to Choose Zephyr Toolchain over Arm Bare Metal When Possible

  #12077 opened Jun 27, 2025

• Plumbing for Arm Example Runnner to use Zephyr Toolchain

  #12078 opened Jun 27, 2025

• [wip] Introduce MergedDataMap

  #12087 opened Jun 28, 2025

• [wip] Add MergedDataMap to method

  #12088 opened Jun 28, 2025

• Add `required_for_abi_source_only = True` to extension/android:executorch_llama

  #12093 opened Jun 28, 2025

• Fix the build broken of examples/llm_manual used in llm/getting-started.md

  #12095 opened Jun 28, 2025

8 Issues closed by 6 people

• [CoreML] Quantization example with the mobilenet_v2 does not work

  #12059 closed Jun 29, 2025

• [Benchmark Issue]IPHONE MAX PRO 15 failure investigation [May 22 2025]

  #11083 closed Jun 27, 2025

• Error Code 32 on module.load_method("forward")

  #12053 closed Jun 27, 2025

• Implement load_into for the file descriptor data loader

  #11563 closed Jun 27, 2025

• _load_for_executorch pybinding cannot load model for training

  #4908 closed Jun 27, 2025

• [exir] MemoryPlanning Verifier assumes that if a model has a user input it has at least 1 tensor input.

  #10522 closed Jun 26, 2025

• Add dtype selective build for OSS

  #9983 closed Jun 26, 2025

• Enable selective build model API in OSS

  #11434 closed Jun 26, 2025

4 Issues opened by 3 people

• Add XNNPack recipes to streamline model transformation pipeline with ExportRecipe

  #12069 opened Jun 27, 2025

• Update documentation for cmake selective build features

  #12065 opened Jun 27, 2025

• [et/devtool] profiling and debugging in one call

  #12064 opened Jun 27, 2025

• [et/devtool] op-level numeric discrepancy detection

  #12063 opened Jun 27, 2025

48 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Arm Zephyr cmake Preset

  #11923 commented on Jun 27, 2025 • 9 new comments

• Introducing NXP Neutron runtime

  #10563 commented on Jun 27, 2025 • 2 new comments

• Milestone2.2: Optimize transposes in XNNPACK partition by removing redundant to_copy ops

  #11316 commented on Jun 28, 2025 • 2 new comments

• Move transformer base definition to extension/llm

  #11917 commented on Jun 27, 2025 • 1 new comment

• Arm backend: Make per-channel quantization default

  #11873 commented on Jun 27, 2025 • 1 new comment

• run mtk_build_examples.sh error

  #11999 commented on Jun 26, 2025 • 0 new comments

• GA models benchmark script

  #11039 commented on Jun 27, 2025 • 0 new comments

• Qualcomm AI Engine Direct - CI for Non-LLM GA model

  #11357 commented on Jun 27, 2025 • 0 new comments

• Phi3 runner uses TextLLMRunner

  #11551 commented on Jun 26, 2025 • 0 new comments

• Enable kleidiAI in llama runner

  #11658 commented on Jun 26, 2025 • 0 new comments

• Implement load_into for SharedPtrDataLoader and add test (#11562)

  #11707 commented on Jun 28, 2025 • 0 new comments

• NXP backend: Add quantization of aten.view

  #11784 commented on Jun 27, 2025 • 0 new comments

• Add last_token_pos in llama_transformer

  #11793 commented on Jun 26, 2025 • 0 new comments

• [XNNPACK] Add support for Linear fused BatchNorm

  #11805 commented on Jun 29, 2025 • 0 new comments

• Arm backend: Move Ethos-U backend to generate TOSA-1.0

  #11852 commented on Jun 28, 2025 • 0 new comments

• Improve memory planning for submodule hierarchies.

  #11860 commented on Jun 28, 2025 • 0 new comments

• [XNNPACK Quantizer] Select between TConvs and Convs

  #11863 commented on Jun 27, 2025 • 0 new comments

• Split neutron backend test based on executor dependency

  #11934 commented on Jun 27, 2025 • 0 new comments

• Fix the math in the FuseMulTensorIntoQuantPass.

  #11946 commented on Jun 26, 2025 • 0 new comments

• [Backend Tester] Add FACTO operator test skeleton

  #11953 commented on Jun 27, 2025 • 0 new comments

• [Backend Tester] Add compliance suite skeleton and operator tests

  #11960 commented on Jun 26, 2025 • 0 new comments

• Arm backend: Add VIT test

  #11996 commented on Jun 27, 2025 • 0 new comments

• [et] generate debug handle before opeartor decomposition

  #11997 commented on Jun 28, 2025 • 0 new comments

• [ET-VK][Ops] linear_qta8a_qga4w_qta8o test framework

  #12005 commented on Jun 26, 2025 • 0 new comments

• [ET-VK][Ops] linear_qta8a_qga4w_qta8o impl and shaders

  #12006 commented on Jun 26, 2025 • 0 new comments

• sym_max support in ExecuTorch

  #11988 commented on Jun 26, 2025 • 0 new comments

• Unified, Configurable AOT Logging

  #11986 commented on Jun 27, 2025 • 0 new comments

• (release/0.5) ./install_requirements.sh fails due to conflicting dependencies.

  #9307 commented on Jun 27, 2025 • 0 new comments

• Multiple Entrypoints and Shared State implementation; Request for comments/feedback.

  #8030 commented on Jun 27, 2025 • 0 new comments

• Add test to ensure scalar_type.fbs remains in sync across the executorch repo

  #11572 commented on Jun 27, 2025 • 0 new comments

• Freeing memory using free() which was not allocated by malloc()

  #11395 commented on Jun 27, 2025 • 0 new comments

• Fail to allocate temp memory by exporting torch.topk

  #8700 commented on Jun 27, 2025 • 0 new comments

• torch.split fails in to_edge

  #11723 commented on Jun 27, 2025 • 0 new comments

• Python apis for loading and saving .ptd from dictionaries

  #8542 commented on Jun 27, 2025 • 0 new comments

• Manual kernel registration to include library names in API

  #11221 commented on Jun 27, 2025 • 0 new comments

• Failed build of apple-ios example

  #11753 commented on Jun 27, 2025 • 0 new comments

• Issues during CoreML GA model testing

  #11542 commented on Jun 27, 2025 • 0 new comments

• Support higher-order while_loop and scan torch ops

  #8769 commented on Jun 28, 2025 • 0 new comments

• fix spec_prop_pass

  #7974 commented on Jun 27, 2025 • 0 new comments

• Fix comment in memory_planning.py

  #8010 commented on Jun 27, 2025 • 0 new comments

• remove the exec_aten namespace

  #8018 commented on Jun 27, 2025 • 0 new comments

• Revert to use mean_out than mean_dim_out

  #8021 commented on Jun 27, 2025 • 0 new comments

• [docs][ez] Fix doc build workflow

  #8079 commented on Jun 27, 2025 • 0 new comments

• Adjust tolerance for quantized XNN conv1d tests

  #8093 commented on Jun 27, 2025 • 0 new comments

• [devtool] create stream_data_sink

  #8604 commented on Jun 27, 2025 • 0 new comments

• [Example] Yolo12 Detection sample with OpenVINO/XNNPACK backend

  #10156 commented on Jun 28, 2025 • 0 new comments

• [pytorch hash update] update the pinned pytorch hash

  #10955 commented on Jun 29, 2025 • 0 new comments

• Add torchao kernels to xcframework

  #10963 commented on Jun 29, 2025 • 0 new comments