executorch

executorch copied to clipboard

Create pull_request_template.md

1

### Summary Adds a default pull request template. Follows how PyTorch tracks changelogs for release notes, which is adding "Release notes: " labels, e.g. [this](https://github.com/pytorch/pytorch/pull/137289) PyTorch PR. Fixes #5793 ###...

dvorjackz

Pull in 16x4 kernel for QP8

2

We are pulling in and testing out new 16x4 kleidi kernels, we see some significant performance improvements from this

mcr229

Installing v0.3.0 fails because torchvision 0.19.0 requires torch==2.4.1

5

### 🐛 Describe the bug ### the issue I'm getting the following dependency conflict error when I attempt to install executorch v0.3.0 on my arm64 macos system: ```shell ./install_requirements.sh --pybind...

grisaitis

[DO NOT LAND] Try to debug Mac CI failure

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #5844

SS-JIA

Adding model stats to aot_arm_compiler

1

* Adds GenericModelEvaluator, which gathers metrics applicable to all models * Adds --evaluate option to enable gathering quantization metrics Signed-off-by: Tom Allsop Change-Id: Ia9b591841f188870fa5e62d0568169498301393d

tom-arm

Unable to locate llama3.2 model and tokenizer files in the ios demo app

1

Following this tutorial: https://github.com/pytorch/executorch/blob/main/examples/demo-apps/apple_ios/LLaMA/docs/delegates/xnnpack_README.md (on main, commit `92d1d1e410b11945869472e88a2247305921989a`, iPhone 12 Pro, iOS 17.6.1) When I drag and drop the model and tokenizer files in finder, they are not visible in...

hietalajulius

Dont quantize the current token for attention

7

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #5715 * #5670 Differential Revision: [D63497872](https://our.internmc.facebook.com/intern/diff/D63497872/)

kimishpatel

[Executorch][quant] Optimize per channel dequantize

9

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #5715 * __->__ #5670 When using quantized kv cache, dequantization routine takes significantly long. This diff just vectorizes dequant per channel for common...

kimishpatel

Add Arm pass-utils and update passes

3

Add pass utils and update passes to make them more concise. Also fix a bug in ConvertSplitToSlicePass, where getitem nodes were not removed. Change-Id: I9346abae93d58bfcbcf727c279c06e58bc94ba6b

Erik-Lundell

Add div decomposition in ArmQuantizer

8

Suggestion or what a pass to decompose a div in ArmQuantizer might look like.

Erik-Lundell