150 Pull requests merged by 43 people

• Arm backend: fix test_llama.py

  #9575 merged Mar 26, 2025

• Update Qualcomm SoCs support list

  #9592 merged Mar 26, 2025

• 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

• support mimi model export

  #9522 merged Mar 22, 2025

• Make export llama checkpoint and param optional

  #9456 merged Mar 22, 2025

• Fix xnnpack quantization discrepancy for non-fp32

  #8488 merged Mar 22, 2025

• Enable optimized op_le broadcast against 1 element tensor

  #9507 merged Mar 22, 2025

• Use gradle to build the AAR

  #9382 merged Mar 22, 2025

• Fix * 4 debug fix that got accidentally landed in op_sdpa.cpp

  #9492 merged Mar 22, 2025

• Use xplat/caffe2:ovrsource_aten_Config.h for arvr build modes

  #9462 merged Mar 22, 2025

• Remove tuple from module implementation

  #9519 merged Mar 22, 2025

• Export Mimi to xnnpack

  #9303 merged Mar 21, 2025

• [ExecuTorch] Add //examples/portable/executor_runner:executor_runner_opt

  #9508 merged Mar 21, 2025

• [ExecuTorch] Add //examples/portable/executor_runner:executor_runner_opt

  #9291 merged Mar 21, 2025

• Handle value and tensor errors as NSErrors keeping the underlying error codes

  #9503 merged Mar 21, 2025

• [Android] Use new Llm package API

  #9495 merged Mar 21, 2025

• Remove Per-Op mode from DQPartitioner

  #9502 merged Mar 21, 2025

• [backend_api] Delete partitioner tags after lowering

  #9501 merged Mar 21, 2025

• Fix lint violation from #9073

  #9496 merged Mar 21, 2025

• Expose ET errors as is

  #9497 merged Mar 21, 2025

• Refactor dtype handling in export_llama

  #9430 merged Mar 21, 2025

• Remove Per-Op mode from DQPartitioner

  #9378 merged Mar 21, 2025

• [backend_api] Delete partitioner tags after lowering

  #9435 merged Mar 21, 2025

• [Android benchmark] Increase cool down time to 180

  #9480 merged Mar 21, 2025

• Arm backend: support int16 and int32 output tables.

  #9359 merged Mar 21, 2025

• Arm backend: Add test pipeline for run_transform_for_annotation_pipeline

  #9488 merged Mar 21, 2025

• Arm backend: Add info about setup_path.sh and arm-none-eabi-gcc

  #9454 merged Mar 21, 2025

• Arm backend: Various fixes to get per channel quantization in place

  #9487 merged Mar 21, 2025

• Migrate java api

  #9478 merged Mar 21, 2025

• Introduce event_tracer_filter_base class

  #9467 merged Mar 21, 2025

• Arm backend: Make setup.sh handle unset EULA variable

  #9446 merged Mar 21, 2025

• [ExecuTorch][XNNPACK] validate quant params before lowering

  #9436 merged Mar 21, 2025

• [Executorch][SDPA] Fix bug in sdpa

  #9479 merged Mar 21, 2025

• CoreML doc updates

  #9463 merged Mar 21, 2025

• Handle case when native_schema is None

  #9377 merged Mar 21, 2025

• Make nightly versions consistent

  #9453 merged Mar 21, 2025

• Fix pyre issues from non-mypy dirs

  #9477 merged Mar 21, 2025

• Upgrade cmake on CI

  #9471 merged Mar 21, 2025

• Run android-perf more frequently

  #9475 merged Mar 20, 2025

• Link quantized ops lib in xnn_executor_runner

  #9443 merged Mar 20, 2025

• [Executorch][SDPA] Fix bug in sdpa

  #9105 merged Mar 20, 2025

• Add the execution permission after the runner file copied to android device.

  #9073 merged Mar 20, 2025

• Do not require checkpoint in quantize() unless it's gptq

  #9470 merged Mar 20, 2025

• Applied quantization for linear with bias=True in pre_quantization

  #9472 merged Mar 20, 2025

• Bump torchao + add unit tests for torchao kernels

  #9396 merged Mar 20, 2025

• [ET-VK][ez] Make sure vTensor is not referencing the global context

  #9469 merged Mar 20, 2025

• Return result from get_backend_name

  #9465 merged Mar 20, 2025

• [ET-VK][ez] Make sure vTensor is not referencing the global context

  #9458 merged Mar 20, 2025

• Collect Named Data Store at construction

  #9370 merged Mar 20, 2025

• [Android docs] Add maven part

  #9442 merged Mar 20, 2025

• [ET-VK][ez] Fix using a temporary variable when creating ComputePipeline

  #9464 merged Mar 20, 2025

• Fix copy pasta error for the error domain

  #9461 merged Mar 20, 2025

• [ET-VK][ez] Fix using a temporary variable when creating ComputePipeline

  #9405 merged Mar 20, 2025

• Arm backend: Fix flaky test LSTM by bumping allowed atol/rtol slightly

  #9452 merged Mar 20, 2025

• Check for nullptr in MethodMeta::uses_backend

  #9457 merged Mar 20, 2025

• Arm backend: Refactor Ops Tests for AvgPool, Clamp, Clone, Conv1d, Sub

  #9039 merged Mar 20, 2025

• Add Android build-tools;35.0.0

  #9429 merged Mar 20, 2025

• Add Module API to query for the number of methods

  #9441 merged Mar 20, 2025

• MethodMeta API to query of backends number and names

  #9440 merged Mar 20, 2025

• [Cadence] Add scalar cases for binary ops (add, mul, sub, div) on HiFi

  #9411 merged Mar 20, 2025

• Keep LiftedConstant in .pte

  #9202 merged Mar 20, 2025

• [ET-VK] Adding all tensor packing support to split op.

  #9439 merged Mar 20, 2025

• [ET-VK] Adding boolean parameters to add_copy_offset_node to specify index calculation function in copy op's shader.

  #9437 merged Mar 20, 2025

• Arm backend: Add WhyNoPartitionReporter and report rejected nodes

  #8963 merged Mar 20, 2025

• Remove proprietary license

  #9389 merged Mar 20, 2025

• Arm backend: Limit number of build jobs

  #9444 merged Mar 20, 2025

• Arm backend: improve common.parameterize decorator

  #9358 merged Mar 20, 2025

• Add SmolLM (smollm2)

  #9354 merged Mar 20, 2025

• [ET-VK] Adding source_offset processing to copy_packed_dim_offset function.

  #9438 merged Mar 20, 2025

• [ET-VK] Adding all tensor packing support to split op.

  #9345 merged Mar 20, 2025

• [ET-VK] Adding source_offset processing to copy_packed_dim_offset function.

  #9344 merged Mar 20, 2025

• [ET-VK] Adding boolean parameters to add_copy_offset_node to specify index calculation function in copy op's shader.

  #9343 merged Mar 20, 2025

• [ExecuTorch][XNNPACK] validate quant params before lowering

  #9299 merged Mar 20, 2025

• Remove protocols and make executorch runtime bridge standalone

  #9398 merged Mar 20, 2025

• [build Folder Migration] Move iOS CI test files

  #9425 merged Mar 20, 2025

• [build Folder Migration] Move wheel files

  #9418 merged Mar 20, 2025

• [build Folder Migration] Move miscellaneous files

  #9417 merged Mar 20, 2025

• fix number of operator error issue

  #9263 merged Mar 19, 2025

• [XNNPACK][Weights Cache] Use sha256 hash of bytes instead of tensor name

  #9413 merged Mar 19, 2025

• install android ndk for mediatek docker

  #9407 merged Mar 19, 2025

• Fix etdump test

  #9410 merged Mar 19, 2025

• [XNNPACK][Weights Cache] Use sha256 hash of bytes instead of tensor name

  #9333 merged Mar 19, 2025

• [trunk.yml] Increase RAM for Arm runners

  #9409 merged Mar 19, 2025

• Add basic ambiguity check in the tests

  #9371 merged Mar 19, 2025

• Create the foundation for testing CoreML export for wheels

  #9381 merged Mar 19, 2025

• Use -fsanitize=address,undefined in debug builds on CI

  #9397 merged Mar 19, 2025

• Remove usage of external flatc in builds and scripts

  #9306 merged Mar 19, 2025

• [ExecuTorch] Arm Ethos:Make get_compile_spec() configurable

  #9403 merged Mar 19, 2025

• [ExecuTorch] Arm Ethos: Add Linear operator tests

  #9402 merged Mar 19, 2025

• [ExecuTorch] Arm Ethos: Do not depend on torch.testing._internal

  #9401 merged Mar 19, 2025

• [ExecuTorch] Arm Ethos:Make get_compile_spec() configurable

  #9323 merged Mar 19, 2025

• [ExecuTorch] Arm Ethos: Add Linear operator tests

  #9322 merged Mar 19, 2025

• use datasink in example runner

  #8920 merged Mar 19, 2025

• [ExecuTorch] Arm Ethos: Do not depend on torch.testing._internal

  #9321 merged Mar 19, 2025

• Support MPS Backend also on iOS < 16 (#9089)

  #9095 merged Mar 19, 2025

• [ET-VK] Adding all tensor packing support to cat op.

  #9380 merged Mar 19, 2025

• Arm backend: Add initial Llama model test case

  #8679 merged Mar 19, 2025

64 Pull requests opened by 32 people

• Alternative approach to handling memory offset shift

  #9406 opened Mar 19, 2025

• Reset IO for the multiple iters output accurancy

  #9412 opened Mar 19, 2025

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

  #9414 opened Mar 19, 2025

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

  #9415 opened Mar 19, 2025

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

  #9416 opened Mar 19, 2025

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

  #9428 opened Mar 19, 2025

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

  #9431 opened Mar 20, 2025

• Add vectorization in elementwise_util (not working yet)

  #9432 opened Mar 20, 2025

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

  #9448 opened Mar 20, 2025

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

  #9450 opened Mar 20, 2025

• Arm backend: Tosa tools update

  #9451 opened Mar 20, 2025

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

  #9466 opened Mar 20, 2025

• [ET-VK][benchmarking][ez] Don't perform copies when benchmarking

  #9468 opened Mar 20, 2025

• Revert "Arm backend: Add initial Llama model test case"

  #9473 opened Mar 20, 2025

• Enable CoreML by default on macOS wheel builds

  #9483 opened Mar 21, 2025

• Arm backend: Refactor any, bitwise, logical tests

  #9499 opened Mar 21, 2025

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

  #9509 opened Mar 21, 2025

• Remove NTSTATUS cast

  #9510 opened Mar 21, 2025

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

  #9511 opened Mar 21, 2025

• Add portable ELU implementation + test

  #9520 opened Mar 22, 2025

• Add optimized ELU implementation

  #9521 opened Mar 22, 2025

• Support channels_last format in portable upsample kernels

  #9526 opened Mar 23, 2025

• 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

• 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

• [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

• Arm backend: Extend convolution support check to 3d

  #9640 opened Mar 26, 2025

• Arm backend: Set qtol=1 for flaky tests.

  #9641 opened Mar 26, 2025

• quantize_ decouple weight quant and dequant output dtypes

  #9642 opened Mar 26, 2025

29 Issues closed by 16 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

• KernelCall failed at instruction 0:16 in operator aten::le.Tensor_out: 0x12

  #9433 closed Mar 22, 2025

• Update Core ML backend docs using backend template

  #8527 closed Mar 21, 2025

• Llama‘s freqs_cos data loss as for convert dtype

  #9393 closed Mar 21, 2025

• torch._dynamo.exc.Unsupported: Unsupported: quantized nyi in meta tensors with fake tensor propagation.

  #8727 closed Mar 21, 2025

• Building from source error

  #9490 closed Mar 21, 2025

• Request of support of aten::__lshift__.Scala and aten::__rshift__.Scala

  #8711 closed Mar 21, 2025

• [Request Help] “torch._dynamo.exc.UserError: Could not guard on data-dependent expression Eq(u77 - 1, 1)"

  #8988 closed Mar 21, 2025

• I want to understand the size of the memory allocated for the forward method on STM32's

  #9447 closed Mar 21, 2025

• The execution permission of qnn_executor_runner file is removed after the runner file copied to android device.

  #9072 closed Mar 20, 2025

• Reproduce/enable DeepSeek R1 Distill Llama 8B

  #7981 closed Mar 20, 2025

• Lifting constants to an external file should not move __lifted_constant_*

  #8809 closed Mar 20, 2025

• [Request Help] I refer the Delegate Debugging document to debug the QLoRA_LLama 3.2-1B model, and want to get the pipeline of linear operation delegation and determine the underlying kernel in XNNPACK

  #9034 closed Mar 20, 2025

• Is Inspector find_total_for_module correct?

  #7200 closed Mar 20, 2025

• example_runner.sh compile error

  #7078 closed Mar 20, 2025

• UNSTABLE trunk / test-arm-backend-delegation / linux-job

  #9399 closed Mar 20, 2025

• UNSTABLE trunk / test-arm-reference-delegation / linux-job

  #9400 closed Mar 20, 2025

• Add SmolLM

  #9324 closed Mar 20, 2025

• Make sure all license headers are BSD license (or compatible)

  #8418 closed Mar 20, 2025

• Fix license headers for MTK backend

  #9279 closed Mar 20, 2025

• Remove the ability to bring your own flatc

  #9145 closed Mar 19, 2025

52 Issues opened by 21 people

• 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 opened Mar 26, 2025

• QNN Export Llama Model Error "failed to load context binary"

  #9637 opened Mar 26, 2025

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

  #9636 opened Mar 26, 2025

• 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

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

  #9525 opened Mar 22, 2025

• CoreML palletization works in PT2E flow

  #9518 opened Mar 21, 2025

• Move Int8DynamicActivationIntxWeightConfig out of torchao/experimental.

  #9516 opened Mar 21, 2025

• Document end-to-end quantization journey in ExecuTorch

  #9515 opened Mar 21, 2025

• Move EmbeddingQuantizer to torchao

  #9514 opened Mar 21, 2025

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

  #9513 opened Mar 21, 2025

• ExecuTorch backends implement PT2E quantizer for PTQ/QAT

  #9512 opened Mar 21, 2025

• CoreML model works with torch.jit.trace, but not torch.export.export

  #9506 opened Mar 21, 2025

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

  #9500 opened Mar 21, 2025

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

  #9491 opened Mar 21, 2025

• Run arm tests in OSS CI unittest-buck

  #9476 opened Mar 20, 2025

• Support exporting QNN models with Python wheels out-of-the-box

  #9474 opened Mar 20, 2025

• Backend Specific Configuration in Runtime

  #9459 opened Mar 20, 2025

• Add HuggingFace tokenizer to executor runner

  #9455 opened Mar 20, 2025

• [etLLM] New config system to export_llama

  #9449 opened Mar 20, 2025

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

  #9445 opened Mar 20, 2025

• Remove EXECUTORCH_BUILD_HOST_TARGETS

  #9404 opened Mar 19, 2025

• Llama 3.2 3B Core ML poor output quality

  #9395 opened Mar 19, 2025

67 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.

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

  #8573 commented on Mar 25, 2025 • 13 new comments

• Deprioritize top level -O2 in CMAKE_CXX_FLAGS_RELEASE

  #9394 commented on Mar 24, 2025 • 5 new comments

• Arm backend: Add tosa_spec info to .tosa files

  #9392 commented on Mar 25, 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

• Arm backend: Remove unused github ci git setup

  #8833 commented on Mar 24, 2025 • 2 new comments

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

  #9266 commented on Mar 26, 2025 • 2 new comments

• Arm backend: Move ReplaceScalarTensorWithFullPass to transforms

  #8998 commented on Mar 24, 2025 • 1 new comment

• Enable xnnpack in aten mode

  #9049 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 export_llama performance regression test using expected ops

  #9158 commented on Mar 26, 2025 • 1 new comment

• Arm backend: Add missing __init__.py to passes

  #8592 commented on Mar 26, 2025 • 1 new comment

• Arm backend: Make passes preserve and update node metadata

  #9362 commented on Mar 21, 2025 • 1 new comment

• Add MediaTek backend CI

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

• Add type error suppressions for upcoming upgrade

  #5352 commented on Mar 21, 2025 • 1 new comment

• Remove tombstone messages proactively

  #8703 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

• Adding new memory plannig algorithm heap_optimized_greedy

  #8773 commented on Mar 21, 2025 • 0 new comments

• Add method variants in dynamic shim to use core tensor/evalue/result types

  #7834 commented on Mar 26, 2025 • 0 new comments

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

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

• Cleanup XNN delegate post removal of upsample decomposition

  #8910 commented on Mar 21, 2025 • 0 new comments

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

  #9356 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

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

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

• Support llama3 training

  #9149 commented on Mar 19, 2025 • 0 new comments

• Arm backend: Add MobileNet v3 testcase

  #9223 commented on Mar 26, 2025 • 0 new comments

• Add support for torch.pow in the Arm backend

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

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

  #9350 commented on Mar 19, 2025 • 0 new comments

• Qualcomm AI Engine Direct - Add submodule quant config setting

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

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

  #9357 commented on Mar 21, 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

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

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

• Implement unary_ufunc functions using elementwise_util

  #9386 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

• Partitioner to not delegate mixed dtype operators

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

• [v0.6.0] Release Tracker

  #9253 commented on Mar 21, 2025 • 0 new comments

• Quantized MobileBert Model Fails with XNNPACK Assertion Error on Android Deployment

  #8994 commented on Mar 21, 2025 • 0 new comments

• Memory Profiling in DevTools

  #8911 commented on Mar 21, 2025 • 0 new comments

• Build pip wheel and validate AOT workflow on different backends & host environments in CI

  #9019 commented on Mar 20, 2025 • 0 new comments

• torch._check_is_size is not being recognized by CoreML partitioner

  #9213 commented on Mar 20, 2025 • 0 new comments

• `torchao:` quantization schemes can't find ao lib

  #8937 commented on Mar 20, 2025 • 0 new comments

• [Android] Flatten data structure like dict in arg types

  #8905 commented on Mar 20, 2025 • 0 new comments

• Vectorize optimized_portable_ops versions of portable ops?

  #9241 commented on Mar 20, 2025 • 0 new comments

• Update MediaTek backend docs using backend template

  #8532 commented on Mar 19, 2025 • 0 new comments

• Update Vulkan backend docs using backend template

  #8529 commented on Mar 19, 2025 • 0 new comments

• Update MPS backend docs using backend template

  #8528 commented on Mar 19, 2025 • 0 new comments

• Update Cadence Xtensa backend docs using backend template

  #8533 commented on Mar 19, 2025 • 0 new comments

• Use real image input in demo colab notebook

  #8851 commented on Mar 19, 2025 • 0 new comments

• Populate "Using ExecuTorch on Android" doc page

  #8536 commented on Mar 19, 2025 • 0 new comments

• Could move some repo resources to S3

  #8805 commented on Mar 19, 2025 • 0 new comments

• [pytorch hash update] update the pinned pytorch hash

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

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

  #9307 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

• [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

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

  #4869 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