54 Pull requests merged by 22 people

• Turn on EXECUTORCH_BUILD_PYBIND when implicitly wanted

  #9611 merged Mar 26, 2025

• Revert "Bump HF version"

  #9600 merged Mar 25, 2025

• Enum for Tensor scalar type and shape dynamism.

  #9609 merged Mar 25, 2025

• Enum for Value tag.

  #9608 merged Mar 25, 2025

• Enum for Module load mode.

  #9606 merged Mar 25, 2025

• Utility helpers to convert between std::vector and NSArray.

  #9597 merged Mar 25, 2025

• Rename deduceScalarType to deduceType

  #9604 merged Mar 25, 2025

• Remove usage of environment variables in wheel building

  #9583 merged Mar 25, 2025

• Switch to new ao quant api for 8da4w (#8501)

  #8772 merged Mar 25, 2025

• Reset Dynamo at Setup for all tests

  #9561 merged Mar 25, 2025

• Fix conflicting pytorch nightly versions

  #9593 merged Mar 25, 2025

• Utility helper to extract a value from NSNumber.

  #9595 merged Mar 25, 2025

• Remove Android ExecuTorch demo

  #9554 merged Mar 25, 2025

• Utility helper to deduce scalar type from NSNumber.

  #9552 merged Mar 25, 2025

• Add a test target to SwiftPM package.

  #9557 merged Mar 25, 2025

• Bump HF version

  #9408 merged Mar 25, 2025

• Create intermediate directories in ETCoreMLModelManager

  #9562 merged Mar 25, 2025

• Refactor serialize.py

  #9579 merged Mar 25, 2025

• Create a CoreML inmemoryfs pybinding target

  #9482 merged Mar 25, 2025

• Tiny fixes to make gcc pedantic build happy

  #8933 merged Mar 25, 2025

• Add NamedData to flat_tensor schema

  #9571 merged Mar 25, 2025

• Arm backend: Add ComputeConstantOpsAOT pass

  #9504 merged Mar 25, 2025

• Refactor serialize.py

  #9124 merged Mar 25, 2025

• Add NamedData to flat_tensor schema

  #9123 merged Mar 25, 2025

• Download checkpoints from HuggingFace

  #9538 merged Mar 25, 2025

• Add missing etdump dependency

  #9547 merged Mar 25, 2025

• Make a real buck file for Apple extension to run tests

  #9560 merged Mar 25, 2025

• Rename phi-4-mini to phi_4_mini

  #9539 merged Mar 25, 2025

• Separate CoreML util and inmemoryfs targets to be reusable

  #9481 merged Mar 24, 2025

• Bump torchao pin to enable dynamic shapes in lowbit

  #9555 merged Mar 24, 2025

• Refactoring memory planning to allow running multiple algorithms

  #8440 merged Mar 24, 2025

• Fix exir.load/save to handle named data store map

  #9485 merged Mar 24, 2025

• Bug fix in Ethos runtime backend

  #9517 merged Mar 24, 2025

• Add Phi-4-mini README.md

  #9302 merged Mar 24, 2025

• Fix qwen import typo

  #9535 merged Mar 24, 2025

• Remove dead test

  #9363 merged Mar 24, 2025

• Use cached PyTorch wheels on MacOS jobs

  #9484 merged Mar 24, 2025

• Add SmolLM (smollm2)

  #9541 merged Mar 24, 2025

• Android aar update and enable emulator test

  #9524 merged Mar 24, 2025

• Updates torchao pin to enable shared embedding quantization

  #9548 merged Mar 24, 2025

• Revert "Add SmolLM (smollm2) (#9354)"

  #9540 merged Mar 24, 2025

• Add rules to build internal code.

  #9546 merged Mar 24, 2025

• Arm backend: Add CEIL Operator

  #9267 merged Mar 24, 2025

• Fix scalar (single element tensor) binary ops on HiFi

  #9523 merged Mar 24, 2025

• Update README to reflect build folder changes

  #9544 merged Mar 24, 2025

• [build Folder Migration] Move core build files into scripts

  #9434 merged Mar 24, 2025

• Add test_mimi to ci

  #9530 merged Mar 24, 2025

• Run setup-macos.sh in unittest-macos even in Buck mode

  #9505 merged Mar 24, 2025

• Arm backend: Fix flaky sigmoid 16/32 bit tests

  #9537 merged Mar 24, 2025

• Arm backend: Add all ops not supported on Ethos-U55 to support-check

  #9489 merged Mar 24, 2025

• updte the return type of log_delegation_intermediate_output

  #9493 merged Mar 24, 2025

• Test mimi: remove redundant codes

  #9528 merged Mar 23, 2025

• test_mimi: Better proxy codes to make it work for both OSS and internal

  #9529 merged Mar 23, 2025

• Qualcomm AI Engine Direct - LPBQ enablement

  #9313 merged Mar 23, 2025

41 Pull requests opened by 24 people

• Added support for bias in optimized linear operation

  #9527 opened Mar 23, 2025

• [ET-VK][ez] Fix 8 bit linear compute shader dispatch

  #9531 opened Mar 23, 2025

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

  #9532 opened Mar 24, 2025

• [WIP]Qualcomm AI Engine Direct - Flags for CI

  #9536 opened Mar 24, 2025

• Refactor LLMEdgeManager's to_dtype

  #9542 opened Mar 24, 2025

• Arm backend: Add ethos_u_minimal_example jupyter notebook

  #9543 opened Mar 24, 2025

• STASH: randn_out

  #9553 opened Mar 24, 2025

• fix rsqrt test case

  #9556 opened Mar 24, 2025

• refactor _Dim into Dim

  #9559 opened Mar 24, 2025

• Fix instrumentation test

  #9564 opened Mar 25, 2025

• Cleaner home page for docs

  #9565 opened Mar 25, 2025

• Make a real buck file for Apple extension to run tests

  #9566 opened Mar 25, 2025

• Arm backend: Size Test for Cortex-M

  #9569 opened Mar 25, 2025

• Qualcomm AI Engine Direct - Mimi Enablement Stage 1

  #9570 opened Mar 25, 2025

• Refactor serialize.py

  #9572 opened Mar 25, 2025

• Arm backend: Clean up shift support

  #9573 opened Mar 25, 2025

• Added hardtanh operator

  #9574 opened Mar 25, 2025

• Arm backend: fix test_llama.py

  #9575 opened Mar 25, 2025

• add sample input for new core op elu.out

  #9577 opened Mar 25, 2025

• Bump PyTorch pin to 2025-03-25

  #9578 opened Mar 25, 2025

• Add SupportedTensorDtypes::{BOOL,REALH}

  #9584 opened Mar 25, 2025

• relax tolerances for all unary float ops

  #9585 opened Mar 25, 2025

• NOCOMMIT: PoC: vectorize just expm1

  #9586 opened Mar 25, 2025

• [TEST] Try to see if sccache is in use

  #9587 opened Mar 25, 2025

• [BE][ET-VK] update permute to use layout gen

  #9588 opened Mar 25, 2025

• [BE][ET-VK] update clone to use layout gen

  #9589 opened Mar 25, 2025

• [BE][ET-VK] update full to use layout gen

  #9590 opened Mar 25, 2025

• [BE][ET-VK] update max_pool2d to use layout gen

  #9591 opened Mar 25, 2025

• Update Qualcomm SoCs support list

  #9592 opened Mar 25, 2025

• [Executorch][llama] Change runner to decouple prompt length from sequence length

  #9594 opened Mar 25, 2025

• Update test_ios_ci.sh to use the app from executorch-examples repo

  #9596 opened Mar 25, 2025

• Add bf16 support to unary_ufunc_realh

  #9599 opened Mar 25, 2025

• Update docs after moving Android ExecuTorchDemo app

  #9601 opened Mar 25, 2025

• Add buck target for hf_download

  #9603 opened Mar 25, 2025

• TEST ONLY

  #9605 opened Mar 25, 2025

• Add mixed dtype check for XNNPACK partitioner

  #9612 opened Mar 25, 2025

• elementwise_util: s/common/compute/ almost everywhere and deprecate SAME_AS_COMPUTE

  #9613 opened Mar 25, 2025

• Deprecate non-internal elementwise_util APIs

  #9621 opened Mar 25, 2025

• Strip size_test binaries and report their sizes in build_size_test.sh

  #9633 opened Mar 25, 2025

• Aot compiler fix

  #9634 opened Mar 25, 2025

• RFC: Compute only in int32/long/float/double for portable ops to save size

  #9635 opened Mar 25, 2025

9 Issues closed by 7 people

• exir.load/exir.save is broken after weight sharing (named data store) addition to EdgeProgramManager

  #9365 closed Mar 25, 2025

• Using Hugging Face model card name in export_llama

  #8872 closed Mar 25, 2025

• [Request help] torch._dynamo.exc.TorchRuntimeError about indices and SymInt

  #9486 closed Mar 25, 2025

• Refactor memory planning to support running multiple algorithms and choose the one that has the best output results.

  #8466 closed Mar 24, 2025

• API reference for Android Java bindings

  #8262 closed Mar 24, 2025

• iOS Executorch app support ANE acceleration ?

  #9352 closed Mar 24, 2025

• Problems when running llava on RPi5 ("Failed to load data for backend XnnpackBackend")

  #9090 closed Mar 24, 2025

• Rename top-level build folder to simplify wheel building

  #9117 closed Mar 24, 2025

• BF16 quantization discrepancy

  #8685 closed Mar 24, 2025

31 Issues opened by 13 people

• Support enumerated shapes for models with multiple inputs

  #9632 opened Mar 25, 2025

• Improve CoreML model loading time

  #9631 opened Mar 25, 2025

• Support better tiling algorithms for ANE

  #9630 opened Mar 25, 2025

• Improve GPU performance

  #9629 opened Mar 25, 2025

• CI to cover different iOS versions

  #9628 opened Mar 25, 2025

• BC/FC guarantees clearly stated

  #9627 opened Mar 25, 2025

• Establish a process for monitoring and resolving CoreML GH issues, including an SLA

  #9626 opened Mar 25, 2025

• Continuous coverage and performance monitoring of GA models for CoreML backend.

  #9625 opened Mar 25, 2025

• Clean up ET CoreML documentation

  #9624 opened Mar 25, 2025

• Support RuntimeSpec

  #9623 opened Mar 25, 2025

• Enumerated shape support

  #9622 opened Mar 25, 2025

• Support dynamic shape in outputBacking

  #9620 opened Mar 25, 2025

• Support reshapeFrequency runtime spec

  #9619 opened Mar 25, 2025

• Dynamic shape support

  #9618 opened Mar 25, 2025

• CoreML quantization is supported by PyTorch flows (PT2E).

  #9617 opened Mar 25, 2025

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

  #9616 opened Mar 25, 2025

• A supported ANE-friendly llama model with good out-of-the-box performance and accuracy.

  #9615 opened Mar 25, 2025

• All GA models are supported by CoreML and have similar performance to using CoreML via torch.export

  #9614 opened Mar 25, 2025

• All GA models export to CoreML via. torch.export with the same performance and user experience as torch.jit.trace

  #9610 opened Mar 25, 2025

• ARM baremetal tests are not running properly

  #9602 opened Mar 25, 2025

• Llava runner job failing after HF transformers update

  #9598 opened Mar 25, 2025

• Beefing up CONTRIBUTING.md to lower the barrier of entry for external contributors

  #9582 opened Mar 25, 2025

• Verify Mimi's accuracy and performance

  #9581 opened Mar 25, 2025

• Export lora llama3b with adapter weights

  #9580 opened Mar 25, 2025

• Missing .so lib for training pybinding

  #9576 opened Mar 25, 2025

• [Request help] Failed running call_function <built-in function getitem>

  #9567 opened Mar 25, 2025

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

  #9558 opened Mar 24, 2025

• Missing try-before-set tests for datasink in etdumpgen tests

  #9551 opened Mar 24, 2025

• Add Android add.pte E2E test

  #9550 opened Mar 24, 2025

• log_intermediate_output_delegate cannot handle error as expected

  #9549 opened Mar 24, 2025

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

  #9534 opened Mar 24, 2025

74 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.

• Support channels_last format in portable upsample kernels

  #9526 commented on Mar 25, 2025 • 4 new comments

• [Proposal] Support Multiple Prefill + Decode in a loop

  #9466 commented on Mar 26, 2025 • 4 new comments

• Serialize NamedDataStoreOutput into PTD.

  #9125 commented on Mar 25, 2025 • 3 new comments

• [Build] use timespec_get to get current time

  #5252 commented on Mar 26, 2025 • 2 new comments

• [etLLM][Config, Part1] Convert Args to DictConfig

  #9450 commented on Mar 24, 2025 • 2 new comments

• Deprioritize top level -O2 in CMAKE_CXX_FLAGS_RELEASE

  #9394 commented on Mar 24, 2025 • 1 new comment

• Use external hf_tokenizer in llama runner

  #9112 commented on Mar 24, 2025 • 1 new comment

• Add MediaTek backend CI

  #7120 commented on Mar 24, 2025 • 1 new comment

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

  #9266 commented on Mar 26, 2025 • 1 new comment

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

  #9388 commented on Mar 25, 2025 • 0 new comments

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

  #9387 commented on Mar 25, 2025 • 0 new comments

• Implement unary_ufunc functions using elementwise_util

  #9386 commented on Mar 25, 2025 • 0 new comments

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

  #9385 commented on Mar 25, 2025 • 0 new comments

• Make PyTorch headers available in optimized_portable_kernels, define ET_USE_PYTORCH_HEADERS

  #9384 commented on Mar 25, 2025 • 0 new comments

• Qualcomm AI Engine Direct - Add submodule quant config setting

  #9355 commented on Mar 24, 2025 • 0 new comments

• Add support for torch.pow in the Arm backend

  #9309 commented on Mar 23, 2025 • 0 new comments

• Arm backend: Add MobileNet v3 testcase

  #9223 commented on Mar 25, 2025 • 0 new comments

• Add export_llama performance regression test using expected ops

  #9158 commented on Mar 25, 2025 • 0 new comments

• Enable xnnpack in aten mode

  #9049 commented on Mar 24, 2025 • 0 new comments

• Arm backend: Added 8 new unit tests for testing various passes.

  #9037 commented on Mar 25, 2025 • 0 new comments

• Arm backend: Move ReplaceScalarTensorWithFullPass to transforms

  #8998 commented on Mar 24, 2025 • 0 new comments

• Add optimized ELU implementation

  #9521 commented on Mar 25, 2025 • 0 new comments

• Add portable ELU implementation + test

  #9520 commented on Mar 25, 2025 • 0 new comments

• Depend on extension/threadpool, not thread_parallel_interface, in buck

  #9511 commented on Mar 25, 2025 • 0 new comments

• Fix -Wglobal-constructors/-Wshadow for executor_runner targets

  #9509 commented on Mar 25, 2025 • 0 new comments

• Arm backend: Refactor any, bitwise, logical tests

  #9499 commented on Mar 24, 2025 • 0 new comments

• Enable CoreML by default on macOS wheel builds

  #9483 commented on Mar 26, 2025 • 0 new comments

• Arm backend: Tosa tools update

  #9451 commented on Mar 25, 2025 • 0 new comments

• Arm Backend: Add New Silu and SDPA Decomp passes to arm_pass_manager

  #9448 commented on Mar 25, 2025 • 0 new comments

• Add vectorization in elementwise_util (not working yet)

  #9432 commented on Mar 25, 2025 • 0 new comments

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

  #9431 commented on Mar 25, 2025 • 0 new comments

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

  #9428 commented on Mar 24, 2025 • 0 new comments

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

  #9416 commented on Mar 24, 2025 • 0 new comments

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

  #9415 commented on Mar 24, 2025 • 0 new comments

• [ET-VK] Adding repeat support to add_copy_packed_dim_offset_node function.

  #9414 commented on Mar 25, 2025 • 0 new comments

• Alternative approach to handling memory offset shift

  #9406 commented on Mar 25, 2025 • 0 new comments

• Arm backend: Add tosa_spec info to .tosa files

  #9392 commented on Mar 25, 2025 • 0 new comments

• Arm backend: Use dbg_fail when node visitors raise exceptions

  #9391 commented on Mar 25, 2025 • 0 new comments

• CoreML palletization works in PT2E flow

  #9518 commented on Mar 24, 2025 • 0 new comments

• double free or corruption (out) on running Llava on x86_64

  #4869 commented on Mar 24, 2025 • 0 new comments

• Move Int8DynamicActivationIntxWeightConfig out of torchao/experimental.

  #9516 commented on Mar 24, 2025 • 0 new comments

• [etLLM] New config system to export_llama

  #9449 commented on Mar 24, 2025 • 0 new comments

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

  #9513 commented on Mar 24, 2025 • 0 new comments

• Document end-to-end quantization journey in ExecuTorch

  #9515 commented on Mar 24, 2025 • 0 new comments

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

  #9307 commented on Mar 24, 2025 • 0 new comments

• I want to understand the allocations made on STM32s by Executorch

  #9491 commented on Mar 24, 2025 • 0 new comments

• Standalone AAR build script

  #8261 commented on Mar 24, 2025 • 0 new comments

• Enable dynamic shapes for torchao:8daXw quant ops

  #8981 commented on Mar 24, 2025 • 0 new comments

• Qwen2.5-0.5b SpinQuant export .pte fail

  #9353 commented on Mar 24, 2025 • 0 new comments

• Remove EXECUTORCH_BUILD_HOST_TARGETS

  #9404 commented on Mar 24, 2025 • 0 new comments

• ExecuTorch backends implement PT2E quantizer for PTQ/QAT

  #9512 commented on Mar 24, 2025 • 0 new comments

• Move EmbeddingQuantizer to torchao

  #9514 commented on Mar 24, 2025 • 0 new comments

• ExecuTorch Scalar to() supports fewer types than c10::Scalar, breaking source compatibility

  #9500 commented on Mar 24, 2025 • 0 new comments

• Typo in import path of examples/models/qwen2_5/__init__.py

  #9525 commented on Mar 24, 2025 • 0 new comments

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

  #9356 commented on Mar 24, 2025 • 0 new comments

• Partitioner to not delegate mixed dtype operators

  #9023 commented on Mar 23, 2025 • 0 new comments

• Qualcomm AI Engine Direct - Support Qnn IR backend in online preparation

  #8876 commented on Mar 24, 2025 • 0 new comments

• Arm backend: Remove unused github ci git setup

  #8833 commented on Mar 24, 2025 • 0 new comments

• Remove tombstone messages proactively

  #8703 commented on Mar 25, 2025 • 0 new comments

• Arm backend: Add missing __init__.py to passes

  #8592 commented on Mar 25, 2025 • 0 new comments

• Openvino backend for Executorch to enable inference on Intel CPUs, GPUs, NPUs

  #8573 commented on Mar 25, 2025 • 0 new comments

• [pytorch hash update] update the pinned pytorch hash

  #4589 commented on Mar 26, 2025 • 0 new comments

• Is the QNN backend support the model of Llama 3.2 3B instead of XNNPACK？

  #9311 commented on Mar 26, 2025 • 0 new comments

• Refactor binary op partitioner configs under binary op config class

  #9024 commented on Mar 26, 2025 • 0 new comments

• QNN GPU or DSP backend issue

  #5914 commented on Mar 25, 2025 • 0 new comments

• [RFC] Canonicalize python file structure

  #8699 commented on Mar 25, 2025 • 0 new comments

• ./install_executorch.sh fails on pip due to dependency conflicts

  #9445 commented on Mar 25, 2025 • 0 new comments

• [Android] Error running a transformer decoder-based neural network

  #8665 commented on Mar 25, 2025 • 0 new comments

• Error in pip editable mode in export_llama

  #9278 commented on Mar 25, 2025 • 0 new comments

• ConstraintViolationError

  #6782 commented on Mar 25, 2025 • 0 new comments

• Parallelize portable ops if threadpool is available, with fallback to parallel_for-as-for-loop

  #8932 commented on Mar 25, 2025 • 0 new comments

• QNN: mobilebert failed to generate Qnn context binary

  #7946 commented on Mar 25, 2025 • 0 new comments

• SM8750 Htp not added in Android apk building

  #8454 commented on Mar 25, 2025 • 0 new comments

• Recipe and Input class definition for executorch.export API

  #9366 commented on Mar 25, 2025 • 0 new comments