Insights: pytorch/executorch
October 15, 2025 – October 22, 2025
Overview
Summary
Excluding merges, 55 authors have pushed 143 commits to main and 522 commits to all branches.
On main, 509 files have changed and there have been 23,415 additions and 6,847 deletions
Top Committers
1 release published by 1 person
- v1.0.0 v1.0.0
Oct 18, 2025
173 pull requests merged by 44 people
- Improving quantized matmul performance by devectorizing shader.
#15274 merged
Oct 22, 2025 - Metal backend: Add MPSGraph caching
#15346 merged
Oct 22, 2025 - Metal backend: eliminate memory leak
#15343 merged
Oct 22, 2025 - Metal backend: track tensors
#15342 merged
Oct 22, 2025 - Metal backend: store weights outside .so
#15341 merged
Oct 22, 2025 - Arm backend: Remove `# type: ignore` from tosa_serializer import
#15301 merged
Oct 22, 2025 - Add desktop README
#15347 merged
Oct 22, 2025 - Arm backend: Improve error logging when running Corstone FVP
#15276 merged
Oct 22, 2025 - Enable int16 for op permute
#15256 merged
Oct 22, 2025 - Llm runner msvc
#15250 merged
Oct 22, 2025 - Msvc ops changes
#15226 merged
Oct 22, 2025 - pin bump tokenizers
#15330 merged
Oct 22, 2025 - reformat inspector doc for better demo (#15325)
#15333 merged
Oct 22, 2025 - Cadence ops: Support quantized gru
#15209 merged
Oct 22, 2025 - patch global cmake build errors
#15211 merged
Oct 22, 2025 - Minor perf improvements to quantized mat mul shader.
#15261 merged
Oct 22, 2025 - Bulk cherry-pick doc updates
#15338 merged
Oct 22, 2025 - Remove extra demo line from sucess-stories page
#15337 merged
Oct 22, 2025 - Minor doc fixes
#15336 merged
Oct 22, 2025 - Fix text_llm_runner kv cache pos count and use it for generate()
#15295 merged
Oct 22, 2025 - Relax verification on flat tensor
#15334 merged
Oct 21, 2025 - Runner logs
#15279 merged
Oct 21, 2025 - Relax verification on flat tensor
#15313 merged
Oct 21, 2025 - Fix more typos and broken links
#15331 merged
Oct 21, 2025 - Update build from source and getting started docs
#15311 merged
Oct 21, 2025 - Handle dataPath = null case
#15326 merged
Oct 21, 2025 - [ET-VK] Contiguous buffer implementation for embedding
#15320 merged
Oct 21, 2025 - Add gemma to supported models
#15328 merged
Oct 21, 2025 - [ET-VK] Implementation of `pow.Tensor_Scalar`
#15319 merged
Oct 21, 2025 - [ET-VK][ez] Introduce a graph config setting to force resize functions to execute
#15318 merged
Oct 21, 2025 - [ET-VK] Introduce `TextureMetadata` struct
#15317 merged
Oct 21, 2025 - gemma3 e2e runner on cuda
#15323 merged
Oct 21, 2025 - add module level benchmark for gemma3 model
#15322 merged
Oct 21, 2025 - [ET-VK][ez] Make pipeline executable properties be controlled by a different macro
#15315 merged
Oct 21, 2025 - make aoti_torch_empty_strided support creating incontiguous tensor
#15321 merged
Oct 21, 2025 - Update success-stories.md
#15309 merged
Oct 21, 2025 - [ET-VK][ez] Accept `sample_kwargs` as an argument in several test util functions
#15314 merged
Oct 21, 2025 - gemma3 e2e runner on cuda
#15282 merged
Oct 21, 2025 - add module level benchmark for gemma3 model
#15241 merged
Oct 21, 2025 - make aoti_torch_empty_strided support creating incontiguous tensor
#15228 merged
Oct 21, 2025 - [ET-VK] Contiguous buffer implementation for embedding
#15160 merged
Oct 21, 2025 - [ET-VK] Implementation of `pow.Tensor_Scalar`
#15159 merged
Oct 21, 2025 - [ET-VK][ez] Introduce a graph config setting to force resize functions to execute
#15158 merged
Oct 21, 2025 - [ET-VK] Introduce `TextureMetadata` struct
#15157 merged
Oct 21, 2025 - [ET-VK][ez] Make pipeline executable properties be controlled by a different macro
#15154 merged
Oct 21, 2025 - [ET-VK][ez] Accept `sample_kwargs` as an argument in several test util functions
#15153 merged
Oct 21, 2025 - Use NoThreadPoolGuard in executor_runner if -cpu_threads is set to 1
#15264 merged
Oct 21, 2025 - Use runtime::FunctionRef in threadpool.h itself
#15265 merged
Oct 21, 2025 - Cadence ops: Support strongly typed softmax
#15201 merged
Oct 21, 2025 - Add Metal backend documentation to Voxtral README
#15273 merged
Oct 21, 2025 - Update docs on LMM runner Apple API
#15307 merged
Oct 21, 2025 - Revise ExecuTorch documentation for Apple runtime
#15293 merged
Oct 21, 2025 - [ET-VK] Add redirect for backends-vulkan
#15305 merged
Oct 21, 2025 - Arm backend: Reduce conv2d unit test sizes
#15271 merged
Oct 21, 2025 - fix: alert users that darwin is not supported
#15289 merged
Oct 21, 2025 - Arm backend: remove some xpassing xfails
#15268 merged
Oct 21, 2025 - Arm backend: Wrap modules in test_nn_modules
#15297 merged
Oct 21, 2025 - update backend cadence md for branch cut
#15277 merged
Oct 21, 2025 - Cadence ops: Add in singleton tensor default variants
#15199 merged
Oct 21, 2025 - Fix text_llm_runner kv cache pos count and use it for generate()
#15286 merged
Oct 21, 2025 - Export LLMs with Optimum docs
#15062 merged
Oct 21, 2025 - Android Docs: Fix stale backend link (android-samsung-exynos)
#15287 merged
Oct 21, 2025 - [doc] Update quantization instructions for clarity
#15284 merged
Oct 20, 2025 - Fix the llm module e2e test
#15275 merged
Oct 20, 2025 - Cadence ops: Support quantized_w8a32_linear
#15171 merged
Oct 20, 2025 - Update usages of DataLoader::SegmentInfo::Type::External
#15195 merged
Oct 20, 2025 - Update XNNPACK doc structure and add template
#14873 merged
Oct 20, 2025 - Website Polish and Cleanup
#15272 merged
Oct 20, 2025 - Dedup constants in emitter
#15139 merged
Oct 20, 2025 - Update 1.0 to stable in executorch-versions.json
#15280 merged
Oct 20, 2025 - Updated Android doc with proper 1.0.0 backend links to executorch
#15266 merged
Oct 20, 2025 - examples/qwen3: match config with optimum flow
#15239 merged
Oct 20, 2025 - Add Metal backend CI workflow with Voxtral testing
#15233 merged
Oct 20, 2025 - Arm backend: Reduce complexity of get_model_and_inputs_from_name
#15247 merged
Oct 20, 2025 - Arm backend: Fix Mypy error related to _QuantProperty.qspec
#14814 merged
Oct 20, 2025 - Android Documentation Improvements and other fixes
#15260 merged
Oct 20, 2025 - Bump transformers pin to 4.56.1
#15243 merged
Oct 20, 2025 - Fix remaining instances of `for aten in (True, False):`
#13529 merged
Oct 20, 2025 - Converting all uint16 to int in quantized mat mul shader to improve perf.
#15193 merged
Oct 19, 2025 - Doubling tile texel col count for mat mul op to improve performance.
#15192 merged
Oct 19, 2025 - Pin bump tokenizer
#15254 merged
Oct 18, 2025 - Success Stories page initial stage
#15236 merged
Oct 18, 2025 - Add int4mm test to the CUDA CI flow
#15181 merged
Oct 18, 2025 - Removing one shift op from quantized linear shader to improve perf.
#15191 merged
Oct 18, 2025 - Reduce log noise
#15244 merged
Oct 18, 2025 - [ET-VK][docs] Update to the new template
#14996 merged
Oct 18, 2025 - Remove default bos/eos from metadata
#15231 merged
Oct 18, 2025 - Fixing cuda buck
#15229 merged
Oct 17, 2025 - Add @Ignore for e2e tests related with ModuleAdd.pte
#15242 merged
Oct 17, 2025 - Cadence ops: Support for contiguous svd
#15142 merged
Oct 17, 2025 - Android add a new MODEL_TYPE_MULTIMODAL const
#15235 merged
Oct 17, 2025 - Reduce log noise
#15177 merged
Oct 17, 2025 - Update mps docs and fix coreml/mps doc references
#15179 merged
Oct 17, 2025 - audio float API
#15234 merged
Oct 17, 2025 - Update 1.0 docs to stable
#15237 merged
Oct 17, 2025 - Android add a new MODEL_TYPE_MULTIMODAL const
#15230 merged
Oct 17, 2025 - Re land asset management
#14768 merged
Oct 17, 2025 - ET Restrict macro
#15225 merged
Oct 17, 2025 - Xnnpack msvc
#15224 merged
Oct 17, 2025 - Moving scale fetch later for minor improvement.
#15190 merged
Oct 17, 2025 - [aoti-et] Store weights outside of .so
#15180 merged
Oct 17, 2025 - All type-specific quantize/dequantize
#15165 merged
Oct 17, 2025 - Qualcomm backend documentation update
#15204 merged
Oct 17, 2025 - Document Quantizer API and precision options for MTK backend
#15203 merged
Oct 17, 2025 - Summary: Pico2 demo of simple neural network (MNIST)
#15196 merged
Oct 17, 2025 - Core ML doc updates
#15175 merged
Oct 17, 2025 - Refine Apple API usage in README
#15122 merged
Oct 17, 2025 - Added support for Stable Diffusion LCM model
#15075 merged
Oct 17, 2025 - Add fpu embedded target to EthosUBackend
#15202 merged
Oct 17, 2025 - audio float API
#15214 merged
Oct 17, 2025 - Support initializing StaticAttentionIOManager from any module with StaticAttention inside
#15206 merged
Oct 17, 2025 - Pin bump tokenizer
#15217 merged
Oct 17, 2025 - Add Metal backend build system and runtime integration
#15024 merged
Oct 17, 2025 - Arm backend: Enable devtools flag when using et-dump in vkml runner
#15221 merged
Oct 17, 2025 - Metal backend: Add operator implementations
#15023 merged
Oct 17, 2025 - Arm backend: Add TOSA (DW) Conv2d dialect operator
#14843 merged
Oct 17, 2025 - data copy ops
#15164 merged
Oct 17, 2025 - Arm backend: Add support for sigmoid and tanh int16x8
#15101 merged
Oct 17, 2025 - Arm backend: Add SmolLM2-135M to CI model testing
#14722 merged
Oct 17, 2025 - Add Pico2 Tutorials on Raspberry Pi
#15188 merged
Oct 17, 2025 - Extend FuseViewCopyTransform to fuse more views
#14745 merged
Oct 17, 2025 - Metal backend: Implement the AOTI MPS shim
#15022 merged
Oct 17, 2025 - Metal backend: Add AOTI shims for memory management
#15021 merged
Oct 17, 2025 - Add Metal backend core ETMetal runtime.
#15020 merged
Oct 17, 2025 - Cadence ops: Get rid of linalg vector norm
#15140 merged
Oct 17, 2025 - platform layer for windows and linux compatibility in cuda_backend
#15183 merged
Oct 17, 2025 - Add -fexceptions to quantized kernel generated libs
#14962 merged
Oct 16, 2025 - Remove ET_UNWRAP usage
#15200 merged
Oct 16, 2025 - Cadence: Support quantized_w8a32_conv
#15137 merged
Oct 16, 2025 - Revert "Don't assign ANDROID_HOME if set already"
#15205 merged
Oct 16, 2025 - Msvc inline macros
#15197 merged
Oct 16, 2025 - Don't assign ANDROID_HOME if set already
#15198 merged
Oct 16, 2025 - Qualcomm backend documentation update
#15043 merged
Oct 16, 2025 - [Samsung][docs] Update to the new template
#15087 merged
Oct 16, 2025 - Allow pre-commit hook to sync c10 headers between ET and pytorch
#15184 merged
Oct 16, 2025 - Arm backend: Update failures to slice operator for int16x8
#15104 merged
Oct 16, 2025 - Arm backend: Streamline MLSDK deps and standardise setup logging
#15189 merged
Oct 16, 2025 - Choosing `ops_to_preserve` by delegating to pattern
#15121 merged
Oct 16, 2025 - Document Quantizer API and precision options for MTK backend
#15091 merged
Oct 16, 2025 - fix compiler flags for msvc in portable lib
#15185 merged
Oct 16, 2025 - Summary: Pico2 demo of simple neural network (MNIST)
#15186 merged
Oct 16, 2025 - Qualcomm AI Engine Direct - Suite operator fix part 3
#15182 merged
Oct 16, 2025 - Source transform to use static attention
#15176 merged
Oct 16, 2025 - Website: Do first pass on fixing the website
#15187 merged
Oct 16, 2025 - NXP backend: Replace pass to fuse activations functions with joint quantization with activation
#14816 merged
Oct 16, 2025 - The padding of concat is not needed anymore if the inputs are equal.
#15102 merged
Oct 16, 2025 - Add extension_llm_runner library to executorch-config.cmake
#15129 merged
Oct 16, 2025 - [ET-VK] Removing manual unroll in linear shader to improve overall performance.
#15110 merged
Oct 16, 2025 - Use torch 2.9 release packages
#15168 merged
Oct 16, 2025 - Android audio input API
#15169 merged
Oct 16, 2025 - Add lora for mlp and unsloth
#15132 merged
Oct 15, 2025 - [aoti-et] Store symbols from dlopen into AOTIDelegateHandle
#15172 merged
Oct 15, 2025 - Core ML doc updates
#15073 merged
Oct 15, 2025 - Qualcomm AI Engine Direct - fix part of suite model test
#15156 merged
Oct 15, 2025 - Add Llama xnnpack recipe
#15167 merged
Oct 15, 2025 - Android LlmModule add API for normalized image input
#15145 merged
Oct 15, 2025 - Add RaspberryPi Tutorials to deploy & infer llama model
#15152 merged
Oct 15, 2025 - Summary: Add the Cross compilation Script for RPi (4 & 5) for Linux host machine
#15151 merged
Oct 15, 2025 - Cadence: Warning if reference kernels not implemented for registered ops
#15130 merged
Oct 15, 2025 - Android audio input API
#15166 merged
Oct 15, 2025 - Fix max seq length bug
#15141 merged
Oct 15, 2025 - Switch ruamel-yaml default from 0.17.x to 0.18.15
#15119 merged
Oct 15, 2025 - Reduce log noise
#15163 merged
Oct 15, 2025 - add method name to cuda error msg
#15133 merged
Oct 15, 2025 - Extend reinplace pass to select_copy.int
#15136 merged
Oct 15, 2025 - Add RaspberryPi Tutorials to deploy & infer llama model
#15109 merged
Oct 15, 2025 - Summary: Add the Cross compilation Script for RPi (4 & 5) for Linux host machine
#15014 merged
Oct 15, 2025 - Arm backend: Enable running MV2/DL3/Conformer on MLSDK runtime
#15098 merged
Oct 15, 2025 - NXP backend: Switch to the default ExecuTorch graph visualization.
#14297 merged
Oct 15, 2025 - Arm backend: Move rescales from MUL visitor to pass
#15103 merged
Oct 15, 2025
53 pull requests opened by 30 people
- NXP Backend: Add padd to remove unnecessary Quantize/Dequantize nodes.
#15148 opened
Oct 15, 2025 - Arm backend: Enable running MV2/DL3/Conformer on MLSDK runtime
#15150 opened
Oct 15, 2025 - [ET-VK] Introduce specialized implementation for per-row reduction
#15161 opened
Oct 15, 2025 - Qualcomm AI Engine Direct - fix suite op
#15162 opened
Oct 15, 2025 - [CoreML] Add retry logic to database/key-value store
#15170 opened
Oct 15, 2025 - Qualcomm AI Engine Direct - fix part of suite model test
#15173 opened
Oct 15, 2025 - Qualcomm AI Engine Direct - Suite operator fix part 3
#15194 opened
Oct 16, 2025 - disable flatcc 3p for cmake build
#15208 opened
Oct 16, 2025 - [cadence][OSS]patch global cmake build errors
#15210 opened
Oct 16, 2025 - Fix incorrect kernel mappings in Cadence HiFi functions.yaml
#15216 opened
Oct 17, 2025 - Arm backend: Fuse duplicate user ops
#15218 opened
Oct 17, 2025 - NXP Backend: Update documentation to the new scheme
#15219 opened
Oct 17, 2025 - Vulkan backend: Add missing include of cstdint
#15220 opened
Oct 17, 2025 - Remove internal executorch dependency on torchao.quantization.subclass
#15223 opened
Oct 17, 2025 - TEST ONLY
#15227 opened
Oct 17, 2025 - [ET-VK][DO NOT LAND] Experimental smem shader for int8 matmul
#15232 opened
Oct 17, 2025 - Update mps docs and fix coreml/mps doc references
#15240 opened
Oct 17, 2025 - Enable E2E test for Exynos Backend
#15245 opened
Oct 18, 2025 - Pin bump oct17
#15251 opened
Oct 18, 2025 - Llm preset cuda win
#15253 opened
Oct 18, 2025 - Fix typo in tutorial for Arm Corstone-320
#15255 opened
Oct 19, 2025 - Fix conv2d bias INT48 type hint removed by fusing placeholder nods
#15257 opened
Oct 19, 2025 - Fix qnn in android demo app
#15258 opened
Oct 19, 2025 - Use xt macros to capture errors returned by nnlib calls
#15259 opened
Oct 19, 2025 - Remove workaround for ovr_config//os:iphoneos not working in OSS
#15263 opened
Oct 19, 2025 - Arm backend: Support per-channel in TOSA.RESCALE
#15267 opened
Oct 20, 2025 - Arm backend: Use reshape instead of view before edge
#15269 opened
Oct 20, 2025 - NXP backend: Improve `view_copy` delegation
#15270 opened
Oct 20, 2025 - Arm backend: ArmTester support testing with portable ops
#15278 opened
Oct 20, 2025 - Update usages of DataLoader::SegmentInfo::Type::External
#15281 opened
Oct 20, 2025 - Android 1.0 release workflow update
#15288 opened
Oct 20, 2025 - Add support for Vulkan lowering
#15290 opened
Oct 21, 2025 - Modify Hexagon SDK paths
#15291 opened
Oct 21, 2025 - adding cuda memory estimation support
#15294 opened
Oct 21, 2025 - Arm backend: Merge passes that replace scalars
#15298 opened
Oct 21, 2025 - Arm backend: Move rescales from SUM visitor to pass
#15299 opened
Oct 21, 2025 - Arm backend: Propagate node info from quantizer to backend
#15300 opened
Oct 21, 2025 - Arm backend: Deprecate internal models using aot_arm_compiler
#15302 opened
Oct 21, 2025 - Fixing issues with using buffers for quantized linear weights.
#15306 opened
Oct 21, 2025 - Update README with mobile demo app details
#15308 opened
Oct 21, 2025 - Add dependency for RemoveCatFromSliceCopyPass.
#15312 opened
Oct 21, 2025 - Properly record load time
#15327 opened
Oct 21, 2025 - Fix command for calling export_llm API
#15329 opened
Oct 21, 2025 - enable int4 tile ci for gemma3
#15332 opened
Oct 21, 2025 - demo colors
#15335 opened
Oct 21, 2025 - add CI configs for msvc build
#15340 opened
Oct 22, 2025 - Add a whisper runner that works with optimum execturoch
#15345 opened
Oct 22, 2025 - Update success stories description for clarity
#15348 opened
Oct 22, 2025 - add readme for gemma3 cuda
#15349 opened
Oct 22, 2025 - clean unnecessary CMAKE_ARGS="-DEXECUTORCH_BUILD_CUDA=ON"
#15350 opened
Oct 22, 2025 - Arm backend: Remove pyre-unsafe from _passes/
#15351 opened
Oct 22, 2025 - Arm backend: Remove pyre-unsafe from tosa/, vgf/ and ethosu/
#15352 opened
Oct 22, 2025 - Temp disable lora test
#15353 opened
Oct 22, 2025
10 issues closed by 7 people
- Android LLM runner should avoid having deps on examples/models/{llama/llava} code
#10119 closed
Oct 21, 2025 - Support Voxtral on Android
#15238 closed
Oct 21, 2025 - Add support for Gemma 3 270M
#14941 closed
Oct 21, 2025 - v1.0.1: tokenizer pypi package
#15303 closed
Oct 21, 2025 - Update Cadence docs
#15249 closed
Oct 21, 2025 - Building 1.0.0 from source with AppleClang
#15262 closed
Oct 21, 2025 - Update Vulkan backend docs using backend template
#8529 closed
Oct 20, 2025 - Missing Half/Bfloat16 support in portable kernels
#13587 closed
Oct 16, 2025 - libc version is too low for docker to build nightly/release wheel package
#14567 closed
Oct 15, 2025
11 issues opened by 10 people
- Samsung runner build fails with -DEXECUTORCH_BUILD_TESTS=ON
#15310 opened
Oct 21, 2025 - v1.0.1 candidate: Fix text_llm_runner kv cache pos count
#15304 opened
Oct 21, 2025 - Error in minimal example code for export with vulken as backend
#15296 opened
Oct 21, 2025 - v1.0.1 candidate: Cpp runtime cant run nanogpt
#15285 opened
Oct 20, 2025 - v1.0.1 candidate: ExecuTorch 1.0 does not have pytorch_tokenizers
#15283 opened
Oct 20, 2025 - Model support: IBM Granite-4.0-H-Micro
#15248 opened
Oct 18, 2025 - Cannot prepare EfficientNet model for QAT due to 'KeyError: _guards_fn'
#15246 opened
Oct 18, 2025 - How to support custom LLMs with qualcomm backend?
#15222 opened
Oct 17, 2025 - executorch.examples.models.llava.image_util deleted in PR #10794
#15178 opened
Oct 15, 2025 - Document JS/wasm APIs in our documentations
#15149 opened
Oct 15, 2025
59 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.
- NXP backend: Unify quantization function inplementations
#15044 commented on
Oct 20, 2025 • new comments - Add INT16 support to permute operation for TOSA backend
#15138 commented on
Oct 21, 2025 • new comments - Cortex_m backend: Fix add implementation
#15100 commented on
Oct 22, 2025 • new comments - Adding Vulkan support for executorch on IG4A
#13806 commented on
Oct 21, 2025 • new comments - Delete tools/cmake/{buck_util,resolve_buck}.py
#13588 commented on
Oct 21, 2025 • new comments - Let's not expose the underlying Method.
#13543 commented on
Oct 20, 2025 • new comments - Zephyr SDK Version Bump v0.17.2->v0.17.3
#13426 commented on
Oct 20, 2025 • new comments - [Profiling] Add python scripts to generate per-op profiling from etdumps
#13302 commented on
Oct 16, 2025 • new comments - [KleidiAI] Always attempt activation packing
#13232 commented on
Oct 16, 2025 • new comments - Correct the flatc/flatcc compilation for x86 ios simulator
#13080 commented on
Oct 22, 2025 • new comments - Qualcomm AI Engine Direct - fix typo
#13030 commented on
Oct 20, 2025 • new comments - draff diff for pybindings task #12790
#12909 commented on
Oct 16, 2025 • new comments - Use ::std::string_view
#12706 commented on
Oct 18, 2025 • new comments - [Used for running CI] Add libkleidiai.a to build_apple_frameworks
#12568 commented on
Oct 16, 2025 • new comments - fake commit
#12460 commented on
Oct 16, 2025 • new comments - PAL File for Arm BareMetal Executor Runner
#12399 commented on
Oct 21, 2025 • new comments - Request for support of ExecuTorch pip package on linux aarch64
#10651 commented on
Oct 15, 2025 • new comments - Utility function for numerical correctness of edge dialect graphs and reference implementations
#14036 commented on
Oct 21, 2025 • new comments - Testing CI
#14433 commented on
Oct 22, 2025 • new comments - NXP backend: Resolve limitations of uncertain tensor formats.
#14576 commented on
Oct 15, 2025 • new comments - Reuse types in _named_data_store and support tensor layouts
#14667 commented on
Oct 15, 2025 • new comments - Arm backend: Add 6D tensor and pixel shuffle/unshuffle support
#14854 commented on
Oct 21, 2025 • new comments - Arm backend: Add support for floor_divide.default
#14933 commented on
Oct 22, 2025 • new comments - pin bump with better architecture
#14957 commented on
Oct 16, 2025 • new comments - Add Post to backend stage to recipes
#14990 commented on
Oct 20, 2025 • new comments - Bump torch nightly pin to 20251012
#15026 commented on
Oct 15, 2025 • new comments - Qualcomm AI Engine Direct - Support floor_divide with int input in QNN HTP backend
#15120 commented on
Oct 16, 2025 • new comments - AI Fix for: Zeros/Ones Tensor Constructor (Android)
#15126 commented on
Oct 15, 2025 • new comments - Support sine operator on XNNPACK
#15144 commented on
Oct 15, 2025 • new comments - NXP backend: Add support for aten.conv_transpose2d.input
#15146 commented on
Oct 21, 2025 • new comments - Publish Linux ARM64 Wheels to PyPI
#7331 commented on
Oct 15, 2025 • new comments - Build for Android NDK with Vulkan fails
#14984 commented on
Oct 15, 2025 • new comments - SmolVLM encoder not getting delegated to XNNPack
#14987 commented on
Oct 15, 2025 • new comments - qwen2.5 0.5b inference with llama.py is normal, but qwen3 0.6b repeats
#14402 commented on
Oct 16, 2025 • new comments - RFC: Jinja2cpp Support on ExecuTorch
#15147 commented on
Oct 16, 2025 • new comments - Add Tutorials for CortexM / Cortex A (Rpi & Pico)
#14410 commented on
Oct 16, 2025 • new comments - Vulkan embeddings give incorrect outputs
#12231 commented on
Oct 17, 2025 • new comments - [XNNPACK] Yolo12 model quantization
#11523 commented on
Oct 17, 2025 • new comments - Stable diffusion qaihub script fails
#14652 commented on
Oct 20, 2025 • new comments - 16KB pagination support for PyTorch ExecuTorch
#11518 commented on
Oct 20, 2025 • new comments - [v1.0.0] Release Tracker
#14288 commented on
Oct 21, 2025 • new comments - ExecuTorch 0.7.0 - Llama 3.2 and Qwen 2.5 Export Issues
#14810 commented on
Oct 21, 2025 • new comments - Android runtime execution - Execution of method forward failed with status 0x12
#14804 commented on
Oct 21, 2025 • new comments - Is there any way to create timestamps by Whisper?
#14403 commented on
Oct 22, 2025 • new comments - fix spec_prop_pass
#7974 commented on
Oct 21, 2025 • new comments - Fix comment in memory_planning.py
#8010 commented on
Oct 21, 2025 • new comments - remove the exec_aten namespace
#8018 commented on
Oct 21, 2025 • new comments - Revert to use mean_out than mean_dim_out
#8021 commented on
Oct 21, 2025 • new comments - Adjust tolerance for quantized XNN conv1d tests
#8093 commented on
Oct 21, 2025 • new comments - [devtool] create stream_data_sink
#8604 commented on
Oct 21, 2025 • new comments - Bump PT Nightly to 04/08/2025
#9973 commented on
Oct 16, 2025 • new comments - [XNNPACK] torchao is installed by default
#10336 commented on
Oct 16, 2025 • new comments - [pytorch hash update] update the pinned pytorch hash
#10955 commented on
Oct 22, 2025 • new comments - BugFix: Modify TPS metric calculation. Add default cpu threads for hybrid CPU system.
#11063 commented on
Oct 21, 2025 • new comments - [QD8-BF16-QB4] Update XNNPACK flatbuffer with new XNNPACK Datatypes
#11164 commented on
Oct 16, 2025 • new comments - Try some QD8-BF16 Experiments
#11466 commented on
Oct 16, 2025 • new comments - Qualcomm AI Engine Direct - LE support
#12164 commented on
Oct 22, 2025 • new comments - Arm backend: Add dump_delegate_data function
#12334 commented on
Oct 20, 2025 • new comments