75 Pull requests merged by 30 people

• Arm backend: Add ethos_u_minimal_example jupyter notebook

  #9543 merged Apr 4, 2025

• Add optimized ELU implementation

  #9521 merged Apr 4, 2025

• Arm backend: Reduce arm_executor_runner binary size

  #9907 merged Apr 4, 2025

• Bump PyTorch pin to 2025-03-25

  #9578 merged Apr 4, 2025

• Arm backend: Skip instead of returning in Llama test

  #9906 merged Apr 4, 2025

• [Executorch][sdpa] Setup the structure to enable quantized gemms for sdpa

  #9912 merged Apr 4, 2025

• [Executorch][SDPA] Remove slice creation

  #9911 merged Apr 4, 2025

• [Executorch][sdpa] Add accidentaly removed flash attentiona args check

  #9910 merged Apr 4, 2025

• set INTERFACE_LINK_LIBARIES for extension_threadpool

  #9839 merged Apr 4, 2025

• Fix build gating on optimized_portable_kernels

  #9838 merged Apr 4, 2025

• [Executorch][sdpa] Refactor sdpa into impl and op

  #9909 merged Apr 4, 2025

• Added hardtanh operator

  #9574 merged Apr 4, 2025

• Try upgrading to CoreML Tools 8.2

  #9807 merged Apr 4, 2025

• Arm backend: Convert assert to throw ValueError in tosa_backend

  #9905 merged Apr 4, 2025

• Arm backend: Handle memory mode corectly in test_model.py

  #9902 merged Apr 4, 2025

• [Executorch][sdpa] Setup the structure to enable quantized gemms for sdpa

  #9889 merged Apr 4, 2025

• [Executorch][SDPA] Remove slice creation

  #9888 merged Apr 4, 2025

• [Executorch][sdpa] Add accidentaly removed flash attentiona args check

  #9887 merged Apr 4, 2025

• [Executorch][sdpa] Refactor sdpa into impl and op

  #9886 merged Apr 4, 2025

• Arm backend: Add support for Leaky ReLU

  #9903 merged Apr 4, 2025

• Arm backend: Remove use of TOSA_DBG_VERBOSE

  #9901 merged Apr 4, 2025

• Arm backend: Align MobileNetV2 with MobileNetV3 unittest

  #9900 merged Apr 4, 2025

• Arm backend: Convert assert to throw TypeError in op_add

  #9897 merged Apr 4, 2025

• Arm backend: Convert assert to throw ValueError in op_tanh

  #9896 merged Apr 4, 2025

• Arm backend: Add info message to assertions in fold_qdq_with_annotated_qparams_pass

  #9895 merged Apr 4, 2025

• Arm backend: Add ERF operator

  #9836 merged Apr 4, 2025

• Arm backend: Add expected failure for test_torch_functions.py

  #9898 merged Apr 4, 2025

• Update test build instructions in kernels readme

  #9885 merged Apr 4, 2025

• Support QK norm in static attention

  #9879 merged Apr 4, 2025

• forward fix

  #9864 merged Apr 3, 2025

• Create release.yml

  #9884 merged Apr 3, 2025

• RMSNorm support - Executorch

  #9844 merged Apr 3, 2025

• Arm backend: Convert assert to throw ValueError in op_sigmoid

  #9875 merged Apr 3, 2025

• Fix scalar type logging from #9751

  #9845 merged Apr 3, 2025

• Arm backend: Limit number of build jobs

  #9874 merged Apr 3, 2025

• [ET-VK] Replace Uniform buffers with push constants for native layer norm op

  #9872 merged Apr 3, 2025

• [ET-VK] Adding round op support.

  #9871 merged Apr 3, 2025

• [ET-VK] Adding all tensor packing support for native layer norm.

  #9870 merged Apr 3, 2025

• Arm backend: Add where.self

  #9869 merged Apr 3, 2025

• Arm backend: Convert assert to throw TypeError in arm_pass_utils

  #9866 merged Apr 3, 2025

• Arm backend: Improve pre-push hook

  #9873 merged Apr 3, 2025

• [ET-VK] Replace Uniform buffers with push constants for native layer norm op

  #9831 merged Apr 3, 2025

• [ET-VK] Adding round op support.

  #9792 merged Apr 3, 2025

• [ET-VK] Adding all tensor packing support for native layer norm.

  #9532 merged Apr 3, 2025

• Arm backend: Size Test for Cortex-M

  #9569 merged Apr 3, 2025

• Update static attention IO manager to use "smart mask" style update

  #9843 merged Apr 3, 2025

• Support multiple prompts in the runner

  #9817 merged Apr 3, 2025

• Qualcomm AI Engine Direct - Add QNN support for to_edge_transform_and_lower

  #9643 merged Apr 2, 2025

• Update android docs

  #9861 merged Apr 2, 2025

• Update android docs

  #9788 merged Apr 2, 2025

• introduce DelegateDebugIntId

  #9832 merged Apr 2, 2025

• Improve android related docs

  #9860 merged Apr 2, 2025

• convert to .md and clean up content on main docs page

  #9776 merged Apr 2, 2025

• Fix link in using-executorch-building-from-source.md

  #9859 merged Apr 2, 2025

• A short term fix for editable mode install failed to import root level module such as exir

  #9818 merged Apr 2, 2025

• Make PyTorch headers available in optimized_portable_kernels, define ET_USE_PYTORCH_HEADERS

  #9384 merged Apr 2, 2025

• Add README.md for the NXP eIQ Neutron Backend

  #9718 merged Apr 2, 2025

• Make Android Module thread-safe and prevent destruction during inference

  #9833 merged Apr 2, 2025

• [Bug Fix] Fix padding when running in NHWC

  #9816 merged Apr 2, 2025

• Arm backend: Disable test case test_w2l_arm.py::test_w2l_u85_BI

  #9835 merged Apr 2, 2025

• Arm Backend: Fixes related to pytorch updates

  #9834 merged Apr 2, 2025

• Fix Fake bench data type

  #9731 merged Apr 2, 2025

• executorch

  #9661 merged Apr 2, 2025

• [Executorch][custo ops] Add prototype defs for custom op

  #9815 merged Apr 2, 2025

• Add cadence.where_Scalar op

  #9764 merged Apr 2, 2025

• fix qnn export

  #9808 merged Apr 2, 2025

• Allow passing in calibration data to convert_pt2

  #9791 merged Apr 1, 2025

• Change training pybind extension_module linkage to static

  #9809 merged Apr 1, 2025

• [ET-VK] Moving repeat functionality from copy_packed_dim_offset into a separate repeat shader.

  #9814 merged Apr 1, 2025

• [ET-VK] Adding all tensor packing support for repeat op.

  #9813 merged Apr 1, 2025

• [ET-VK] Simplify lane offset copy logic in copy_packed_dim_offset shader.

  #9812 merged Apr 1, 2025

• [ET-VK] Moving repeat functionality from copy_packed_dim_offset into a separate repeat shader.

  #9428 merged Apr 1, 2025

• [ET-VK] Adding all tensor packing support for repeat op.

  #9415 merged Apr 1, 2025

• [ET-VK] Simplify lane offset copy logic in copy_packed_dim_offset shader.

  #9416 merged Apr 1, 2025

• [0.6 Release] Add 0.6 to ET docs index

  #9798 merged Apr 1, 2025

26 Pull requests opened by 19 people

• Expose symbols on macos in the xplat pytorch stack

  #9819 opened Apr 1, 2025

• [ExecuTorch][to_backend] add AllNodePartitioner

  #9822 opened Apr 1, 2025

• [Executorch][to_backend] Introduce preprocess_multimethod

  #9823 opened Apr 1, 2025

• [ExecuTorch][to_backend] Enable to_backend API to leverage preprocess_multimethod

  #9824 opened Apr 1, 2025

• DO NOT COMMIT test for u55 + mv2

  #9830 opened Apr 2, 2025

• Add build_optimized_size_test.sh

  #9840 opened Apr 2, 2025

• Migrate elementwise_util callers to the variants with out_dtypes in template arguments

  #9841 opened Apr 2, 2025

• Save some size in dtype_util when dtype selective build is not in use

  #9842 opened Apr 2, 2025

• [do not land] lora experiment

  #9863 opened Apr 2, 2025

• Add Intel macOS check; update docs

  #9865 opened Apr 3, 2025

• Fix breaks from googletest upgrade (vulkan_compute_api_test) (#9760)

  #9868 opened Apr 3, 2025

• NXP backend: Add NeutronQuantizer

  #9876 opened Apr 3, 2025

• Adding bmm, mm, view_copy, slice_copy, split_with_sizes_copy optimizations

  #9877 opened Apr 3, 2025

• [WIP] Mimi 4-bit quant on transformer and 8-bit on conv

  #9882 opened Apr 3, 2025

• [ET-VK] Improve packing format for int4 linear operator + misc improvements

  #9883 opened Apr 3, 2025

• Reapply "Depend on extension/threadpool, not thread_parallel_interface, in buck (#9511)"

  #9890 opened Apr 3, 2025

• Update release.yaml

  #9891 opened Apr 4, 2025

• [ET-VK] Minor performance improvements to native layer norm.

  #9892 opened Apr 4, 2025

• Arm backend: Add FuseEqualPlaceholdersPass

  #9893 opened Apr 4, 2025

• Add a namespace for ATen mode

  #9894 opened Apr 4, 2025

• Arm backend: Add pre-push checks for op tests

  #9899 opened Apr 4, 2025

• Arm backend: Remove node vistor for full

  #9904 opened Apr 4, 2025

• Arm backend: Add TOSA support for gt.Scalar and lt.Scalar

  #9908 opened Apr 4, 2025

• port hardtanh and add hardtanh test

  #9914 opened Apr 4, 2025

• Fix mobile bert fine tune

  #9915 opened Apr 4, 2025

• export llama with lora

  #9916 opened Apr 4, 2025

5 Issues closed by 3 people

• Migrate backend quantizers/partitioners to use new torchao quant primitives

  #9513 closed Apr 2, 2025

• `from executorch.exir import CaptureConfig` fails with ExecuTorch pip installed in editable mode

  #9558 closed Apr 2, 2025

• Cannot run .pte generated with Python in C++

  #9763 closed Apr 2, 2025

• Missing .so lib for training pybinding

  #9576 closed Apr 1, 2025

• Why does the weight data type of the Linear layer become FP32 during the runtime when load fp16.pte (fp16 Llama 3.2-1B model)?

  #9639 closed Apr 1, 2025

17 Issues opened by 3 people

• Improve contributor documentation for Android

  #9913 opened Apr 4, 2025

• Support BFloat16 dtype in Android Tensor API

  #9881 opened Apr 3, 2025

• UNSTABLE trunk / test-qnn-model (fp32)

  #9878 opened Apr 3, 2025

• [ExecuTorch] recognizes and warns user when Q/DQ ops are not lowered

  #9858 opened Apr 2, 2025

• [torchao] Enable new quant primitives in PT2E

  #9857 opened Apr 2, 2025

• [torchao] Migrate exact copy of PT2E from torch.ao to torchao

  #9856 opened Apr 2, 2025

• [XNNPACK] partitioner recognizes new torchao quant primitives

  #9855 opened Apr 2, 2025

• [QNN] partitioner recognizes new torchao quant primitives

  #9854 opened Apr 2, 2025

• [CoreML] partitioner recognizes new torchao quant primitives

  #9853 opened Apr 2, 2025

• Make ExecuTorch Q/DQ representation default and resilient

  #9852 opened Apr 2, 2025

• Use eager mode quantize_ with QDQLayout for non-delegated quantized ops in ExecuTorch

  #9851 opened Apr 2, 2025

• Migrate PT2E to torchao using the new quant primitives

  #9850 opened Apr 2, 2025

• Standardize ExecuTorch backends on new torchao primitives

  #9849 opened Apr 2, 2025

• [torchao] support codebook/LUT in torchao

  #9848 opened Apr 2, 2025

• [CoreML] quantizer supports 4-bit affine (groupwise and channelwise)

  #9847 opened Apr 2, 2025

• [XNNPACK] support 8d4w in PT2E flow

  #9846 opened Apr 2, 2025

• [0.6 Release] Quality testing

  #9837 opened Apr 2, 2025

49 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.

• Add vectorization in elementwise_util

  #9432 commented on Apr 3, 2025 • 18 new comments

• etdumpfilter impl

  #9752 commented on Apr 4, 2025 • 16 new comments

• [Core ML] Improve error logging

  #9801 commented on Apr 4, 2025 • 1 new comment

• Add code structure and a few other links to CONTRIBUTING.md

  #9793 commented on Apr 3, 2025 • 1 new comment

• NOCOMMIT: PoC: vectorize just expm1

  #9586 commented on Apr 2, 2025 • 1 new comment

• Use external hf_tokenizer in llama runner

  #9112 commented on Apr 2, 2025 • 1 new comment

• Qualcomm AI Engine Direct - Add submodule quant config setting

  #9355 commented on Apr 3, 2025 • 1 new comment

• Recursive initialization of repos

  #9783 commented on Apr 2, 2025 • 0 new comments

• [docs][ez] Fix doc build workflow

  #8079 commented on Apr 2, 2025 • 0 new comments

• Adjust tolerance for quantized XNN conv1d tests

  #8093 commented on Apr 2, 2025 • 0 new comments

• aten.leakyrelu.default Op registery

  #8227 commented on Apr 1, 2025 • 0 new comments

• [devtool] create stream_data_sink

  #8604 commented on Apr 2, 2025 • 0 new comments

• Qualcomm AI Engine Direct - Enable custom operator

  #8726 commented on Apr 4, 2025 • 0 new comments

• Arm backend: Remove unused github ci git setup

  #8833 commented on Apr 3, 2025 • 0 new comments

• Enable xnnpack in aten mode

  #9049 commented on Apr 2, 2025 • 0 new comments

• Qualcomm AI Engine Direct - oss model enablement (EfficientSAM)

  #9266 commented on Apr 2, 2025 • 0 new comments

• Arm backend: Enable mypy lintrunner for backends/arm/test

  #9357 commented on Apr 4, 2025 • 0 new comments

• elementwise_util: don't cast the result of compute_fun back to the common type

  #9385 commented on Apr 2, 2025 • 0 new comments

• Refactor elementwise_util: create variants with out_dtypes in template argument list

  #9387 commented on Apr 2, 2025 • 0 new comments

• RFC: Specialize for non-mixed-dtype in elementwise_util

  #9388 commented on Apr 2, 2025 • 0 new comments

• BroadcastIndexesRange: leading 1s don't require true broadcasting

  #9431 commented on Apr 2, 2025 • 0 new comments

• Arm backend: Tosa tools update

  #9451 commented on Apr 4, 2025 • 0 new comments

• Added support for bias in optimized linear operation

  #9527 commented on Apr 2, 2025 • 0 new comments

• Migrate elementwise_util callers to the variants with out_dtypes in template arguments

  #9741 commented on Apr 2, 2025 • 0 new comments

• Refactor internal switch cases

  #9802 commented on Apr 4, 2025 • 0 new comments

• Model can't inference for Llama3.2-1B when use -d fp16 to convert pte

  #9534 commented on Apr 2, 2025 • 0 new comments

• insert_write_back_for_buffers_pass should inject copy_ nodes at the earliest possible spot.

  #7345 commented on Apr 2, 2025 • 0 new comments

• ExecuTorch backends implement PT2E quantizer for PTQ/QAT

  #9512 commented on Apr 2, 2025 • 0 new comments

• [CoreML] support palletization in PT2E flow

  #9518 commented on Apr 2, 2025 • 0 new comments

• [torchao/ExecuTorch] move embedding quantization to torchao

  #9514 commented on Apr 2, 2025 • 0 new comments

• [v0.6.0] Release Tracker

  #9253 commented on Apr 2, 2025 • 0 new comments

• Label tracking meta-issue (edit me to get automatically CC'ed on issues! cc bot)

  #7679 commented on Apr 2, 2025 • 0 new comments

• [torchao] finalize affine quant primitives

  #9782 commented on Apr 2, 2025 • 0 new comments

• [torchao] Move Int8DynamicActivationIntxWeightConfig out of torchao/experimental.

  #9516 commented on Apr 2, 2025 • 0 new comments

• Binary Size: Selective Build

  #9010 commented on Apr 3, 2025 • 0 new comments

• Android: Tensor type is not very friendly to BFloat16

  #6571 commented on Apr 3, 2025 • 0 new comments

• Don't Serialize Scales/ZP in Flatbuffer

  #9029 commented on Apr 4, 2025 • 0 new comments

• Windows build failure: Failed to query buck for sources.

  #9616 commented on Apr 4, 2025 • 0 new comments

• Return "platform not supported" when using PyTorch on intel-based Macbooks

  #9772 commented on Apr 4, 2025 • 0 new comments

• Facing linking error "undefined reference to `executorch::extension::FileDataLoader::from"

  #9636 commented on Apr 4, 2025 • 0 new comments

• [Request impl] Devtool end-to-end tests

  #9778 commented on Apr 4, 2025 • 0 new comments

• Support Dynamically Quantized Convolutions

  #9021 commented on Apr 4, 2025 • 0 new comments

• Issues with deloyment on RP2040

  #7177 commented on Apr 4, 2025 • 0 new comments

• Query regarding support of Executorch for ARM Ethos-U65 backend

  #9356 commented on Apr 4, 2025 • 0 new comments

• [pytorch hash update] update the pinned pytorch hash

  #4589 commented on Apr 4, 2025 • 0 new comments

• fix spec_prop_pass

  #7974 commented on Apr 2, 2025 • 0 new comments

• Fix comment in memory_planning.py

  #8010 commented on Apr 2, 2025 • 0 new comments

• remove the exec_aten namespace

  #8018 commented on Apr 2, 2025 • 0 new comments

• Revert to use mean_out than mean_dim_out

  #8021 commented on Apr 2, 2025 • 0 new comments