From be73413f797bf6881b1d3bf0f0184db55214da1a Mon Sep 17 00:00:00 2001 From: Puneet Matharu Date: Wed, 3 Jun 2026 16:14:15 +0100 Subject: [PATCH] Remove remaining references to ACL. --- ML-Frameworks/pytorch-aarch64/CHANGELOG.md | 2 +- ML-Frameworks/pytorch-aarch64/README.md | 5 +- ML-Frameworks/pytorch-aarch64/build-wheel.sh | 3 +- .../pytorch-aarch64/examples/README.md | 46 ++++++++----------- ML-Frameworks/pytorch-aarch64/versions.sh | 1 - 5 files changed, 23 insertions(+), 34 deletions(-) diff --git a/ML-Frameworks/pytorch-aarch64/CHANGELOG.md b/ML-Frameworks/pytorch-aarch64/CHANGELOG.md index 8b8e0ba8..4a77882f 100644 --- a/ML-Frameworks/pytorch-aarch64/CHANGELOG.md +++ b/ML-Frameworks/pytorch-aarch64/CHANGELOG.md @@ -45,11 +45,11 @@ where `YY` is the year, and `MM` the month of the increment. - `IDEEP_HASH=e087b6e4b32a7ba684db82231d1558123968ac1d`, from ideep_pytorch, May 11th, 2026. - `ONEDNN_HASH=3004f0a1d9cf92c06eaaca57840aaa2149ebba85`, from main, May 27th, 2026. - `KLEIDIAI_HASH=5866364d3bc079d2d6cae5f0acf6d076594bc7a7`, v1.25.0 from main, May 28th, 2026. - - Replaces `ACL_VERSION=v52.8.0` with `ACL_VERSION=v53.1.0`, from main, May 18th. - Updates `OPENBLAS_VERSION` from `d26960a21ec5da7f77377f28bd6e230060841ae0` to v0.3.33, from main, Apr 23rd. - Updates `torchvision` from 0.26.0.dev20260329 to 0.28.0.dev20260527. ### Removed + - Removes `ACL_VERSION` which references library that is no longer present. - Disables PyTorch [PR #182655](https://github.com/pytorch/pytorch/pull/182655), to update the PyTorch CI build scripts. - Disables PyTorch [PR #170600](https://github.com/pytorch/pytorch/pull/170600), to gate deletion of clean-up steps in build_common.sh. - Disables PyTorch [PR #167328](https://github.com/pytorch/pytorch/pull/167328), to build cpuinfo into c10 shared library. diff --git a/ML-Frameworks/pytorch-aarch64/README.md b/ML-Frameworks/pytorch-aarch64/README.md index dcf3ab32..0aa52b49 100644 --- a/ML-Frameworks/pytorch-aarch64/README.md +++ b/ML-Frameworks/pytorch-aarch64/README.md @@ -109,10 +109,7 @@ Note: use the `CommitDate` for the trailing comments unless otherwise specified #### Tags -For these dependencies, you should assign the latest tag from the releases to the appropriate variable in `versions.sh` (e.g. assign the latest tag for `ComputeLibrary` to `ACL_VERSION`). - -- ComputeLibrary: https://github.com/ARM-software/ComputeLibrary/tags - - Pick the newest release tag. +For these dependencies, you should assign the latest tag from the releases to the appropriate variable in `versions.sh` (e.g. assign the latest tag for `OpenBLAS` to `OPENBLAS_VERSION`). - OpenBLAS: https://github.com/OpenMathLib/OpenBLAS/tags - Pick the newest release tag. diff --git a/ML-Frameworks/pytorch-aarch64/build-wheel.sh b/ML-Frameworks/pytorch-aarch64/build-wheel.sh index 18bb05f4..179f0b28 100755 --- a/ML-Frameworks/pytorch-aarch64/build-wheel.sh +++ b/ML-Frameworks/pytorch-aarch64/build-wheel.sh @@ -94,10 +94,9 @@ mkdir -p "${OUTPUT_LOCAL_DIR}" trap cleanup EXIT -echo "Building local manywheel builder image with ACL_VERSION=${ACL_VERSION} and OPENBLAS_VERSION=${OPENBLAS_VERSION}" +echo "Building local manywheel builder image with OPENBLAS_VERSION=${OPENBLAS_VERSION}" ( cd "${PYTORCH_LOCAL_DIR}" - ACL_VERSION="${ACL_VERSION}" \ OPENBLAS_VERSION="${OPENBLAS_VERSION}" \ MAX_JOBS="${MAX_JOBS}" \ USE_CCACHE="${USE_CCACHE}" \ diff --git a/ML-Frameworks/pytorch-aarch64/examples/README.md b/ML-Frameworks/pytorch-aarch64/examples/README.md index f6c0970a..ad79e55e 100644 --- a/ML-Frameworks/pytorch-aarch64/examples/README.md +++ b/ML-Frameworks/pytorch-aarch64/examples/README.md @@ -1,5 +1,5 @@ @@ -9,25 +9,23 @@ SPDX-License-Identifier: Apache-2.0 -- [Examples](#examples) - - [Description](#description) - - [Vision](#vision) - - [Image classification](#image-classification) - - [Object detection](#object-detection) - - [Natural Language Processing (NLP)](#natural-language-processing-nlp) - - [Question answering](#question-answering) - - [Dynamic quantization](#dynamic-quantization) - - [General optimization guidelines](#general-optimization-guidelines) - - [Weight prepacking](#weight-prepacking) - - [General flags](#general-flags) - - [Compiled mode flags](#compiled-mode-flags) - - [Eager mode flags](#eager-mode-flags) - - [Generative AI](#generative-ai) - - [4 bit Dynamic Quantization](#4-bit-dynamic-quantization) - - [Vision](#vision-1) - - [Command-Line Options](#command-line-options) - - [Text Generation](#text-generation) - - [Command-Line Options](#command-line-options-1) +- [Description](#description) +- [Vision](#vision) + - [Image classification](#image-classification) + - [Object detection](#object-detection) +- [Natural Language Processing (NLP)](#natural-language-processing-nlp) + - [Question answering](#question-answering) +- [Dynamic quantization](#dynamic-quantization) +- [General optimization guidelines](#general-optimization-guidelines) + - [General flags](#general-flags) + - [Compiled mode flags](#compiled-mode-flags) + - [Eager mode flags](#eager-mode-flags) +- [Generative AI](#generative-ai) + - [4 bit Dynamic Quantization](#4-bit-dynamic-quantization) + - [Vision](#vision-1) + - [Command-Line Options](#command-line-options) + - [Text Generation](#text-generation) + - [Command-Line Options](#command-line-options-1) ## Description @@ -56,7 +54,7 @@ The file [`resnet_v1-50.yml`](resnet_v1-50.yml) provides, in [YAML format](https ### Object detection -The script [`detect_objects.py`](detect_object.py) demonstrates how to run object detection using SSD-ResNet-34. +The script [`detect_objects.py`](detect_objects.py) demonstrates how to run object detection using SSD-ResNet-34. The SSD-ResNet-34 model is trained from the Common Object in Context (COCO) image dataset. This is a multiscale SSD (Single Shot Detection) model based on the ResNet-34 backbone network that performs object detection. @@ -198,15 +196,11 @@ Note that in the above data we used the `--warmup` flag to run the model once be ## General optimization guidelines -### Weight prepacking - -`Linear` layers calling [Arm ComputeLibrary](https://github.com/ARM-software/ComputeLibrary) (ACL) matmuls reorder weights during runtime by default. These reorders can be eliminated by calling `pack_linear_weights` as shown in `pack_linear_weights.py`. This improves the performance of any models calling a `Linear` layer multiple times. - ### General flags There are several flags which typically improve the performance of PyTorch. -`DNNL_DEFAULT_FPMATH_MODE`: setting the environment variable `DNNL_DEFAULT_FPMATH_MODE` to `BF16` or `ANY` will instruct ACL to dispatch fp32 workloads to bfloat16 kernels where hardware support permits. _Note: this may introduce a drop in accuracy._ +`DNNL_DEFAULT_FPMATH_MODE`: setting the environment variable `DNNL_DEFAULT_FPMATH_MODE` to `BF16` or `ANY` will result in fp32 workloads being dispatched to bfloat16 kernels where hardware support permits. _Note: this may introduce a drop in accuracy._ You can control the number of threads with `OMP_NUM_THREADS`, smaller models may perform better with fewer threads. diff --git a/ML-Frameworks/pytorch-aarch64/versions.sh b/ML-Frameworks/pytorch-aarch64/versions.sh index 2b22249a..b398eecf 100755 --- a/ML-Frameworks/pytorch-aarch64/versions.sh +++ b/ML-Frameworks/pytorch-aarch64/versions.sh @@ -15,7 +15,6 @@ ONEDNN_HASH=3004f0a1d9cf92c06eaaca57840aaa2149ebba85 # From main, May 27th, 2 KLEIDIAI_HASH=5866364d3bc079d2d6cae5f0acf6d076594bc7a7 # v1.25.0 from main, May 28th, 2026 # build-wheel.sh deps -ACL_VERSION="v53.1.0" # May 18th OPENBLAS_VERSION="v0.3.33" # Apr 23rd # Dockerfile deps