Clean-up + add hygiene checks

.github/workflows/hygiene.yml

@@ -0,0 +1,46 @@
+# SPDX-FileCopyrightText: Copyright 2026 Arm Limited and affiliates.
+#
+# SPDX-License-Identifier: Apache-2.0
+
+name: Hygiene
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    types: [opened, synchronize, reopened]
+  workflow_dispatch:
+
+#* Stop stale workflows when pull requests are updated: https://stackoverflow.com/a/70972844
+#* Does not apply to the main branch.
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+
+# Declare default permissions as read only.
+permissions: read-all
+
+jobs:
+  hygiene:
+    runs-on: ah-ubuntu_24_04-c7g_2x-50
+    steps:
+      - name: Checkout Tool-Solutions
+        uses: actions/checkout@v6.0.2
+
+      - name: Set up Python
+        uses: actions/setup-python@v6.0.0
+        with:
+          python-version: "3.12"
+
+      - name: Install hygiene tooling
+        run: |
+          python -m pip install --user --no-cache-dir pre-commit==4.5.1 reuse==6.2.0
+          echo "$HOME/.local/bin" >> "$GITHUB_PATH"
+
+      # The manual stage contains every whole-repo-safe hook and excludes the
+      # copyright annotation hooks, which intentionally mutate changed files.
+      - name: Run whole-repo pre-commit checks
+        run: pre-commit run --all-files --hook-stage manual
+
+      - name: Run whole-repo REUSE lint
+        run: reuse lint

.pre-commit-config.yaml

@@ -1,13 +1,15 @@
-# SPDX-FileCopyrightText: Copyright 2025 Arm Limited and affiliates.
+# SPDX-FileCopyrightText: Copyright 2025, 2026 Arm Limited and affiliates.
 #
 # SPDX-License-Identifier: Apache-2.0
 
 fail_fast: false
+default_stages: [ pre-commit, manual ]
 
 repos:
   - repo: local
     hooks:
-      # Python/YAML/shell: use hash comments (no --multi-line)
+      # Python/YAML/shell: use hash comments (no --multi-line). This hook
+      # mutates copyright metadata. Do not run this hook with --all-files.
       - id: reuse-annotate-current-year-python-yaml-shell-files
         name: REUSE Annotate (Python add current year, merge)
         language: python
@@ -27,7 +29,8 @@ repos:
         files: '(\.gitignore|\.dockerignore|requirements\.txt|bash_profile|Dockerfile|\.(py|yml|yaml|sh|bash))$'
         stages: [ pre-commit ]
 
-      # JSON/Markdown/welcome.txt: use sidecar .license files
+      # JSON/Markdown/welcome.txt: use sidecar .license files. This hook
+      # mutates copyright metadata. Do not run this hook with --all-files.
       - id: reuse-annotate-current-year-json-and-md
         name: REUSE Annotate (JSON/Markdown/welcome.txt sidecar fallback add current year, merge)
         language: python
@@ -49,12 +52,37 @@ repos:
   - repo: https://github.com/fsfe/reuse-tool
     rev: v6.2.0
     hooks:
+      # File-level REUSE validation for changed files. CI also runs `reuse lint`
+      # across the whole repo because validation is non-mutating.
       - id: reuse-lint-file
-        stages: [ pre-commit ]
         name: "REUSE Compliant Copyright and License"
 
   - repo: https://github.com/pre-commit/pre-commit-hooks
     rev: v6.0.0
     hooks:
       - id: trailing-whitespace
-        stages: [ pre-commit ]
+      - id: end-of-file-fixer
+      - id: mixed-line-ending
+      # Cheap changed-file guard for accidentally committed conflict markers.
+      # GitHub detects active merge conflicts, but this also catches conflict
+      # markers committed into files.
+      - id: check-merge-conflict
+      # Avoid adding paths that collide on case-insensitive filesystems.
+      - id: check-case-conflict
+      # Catch symlinks that point to missing targets. This is normally a no-op
+      # for this repo, but keeps broken links from entering via future changes.
+      - id: check-symlinks
+      - id: check-json
+      - id: check-yaml
+      # Enforce the size limit for every file in whole-repo CI checks as well as
+      # for newly added files in normal pre-commit runs.
+      - id: check-added-large-files
+        args: [ --maxkb=1024, --enforce-all ]
+      - id: check-executables-have-shebangs
+      # Keep script shebangs and executable bits consistent for runnable files.
+      - id: check-shebang-scripts-are-executable
+
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.15.15
+    hooks:
+      - id: ruff-check

CONTRIBUTING.md

@@ -1,5 +1,5 @@
 <!--
-SPDX-FileCopyrightText: Copyright 2025 Arm Limited and affiliates.
+SPDX-FileCopyrightText: Copyright 2025, 2026 Arm Limited and affiliates.
 
 SPDX-License-Identifier: Apache-2.0
 -->
@@ -84,9 +84,19 @@ pre-commit --version
    - Re-run:
 
         ```bash
-        pre-commit run --all-files
+        pre-commit run
         ```
 
+   The command above checks the currently staged files and may update their
+   copyright metadata. Inspect and stage any hook changes before rerunning it.
+
+   To run every repository-safe check across the whole repository without
+   changing copyright metadata, use:
+
+    ```bash
+    pre-commit run --all-files --hook-stage manual
+    ```
+
 5. Push once everything passes:
 
     ```bash

ML-Frameworks/pytorch-aarch64/.agents/skills/bump-sources/SKILL.md

@@ -57,4 +57,4 @@ Provide:
 
 - A short summary of what changed (hashes/tags/wheels).
 - The edits made to `./versions.sh` (and `./get-source.sh` if changed).
-- Commands run and key results/errors.
+- Commands run and key results/errors.

ML-Frameworks/pytorch-aarch64/Dockerfile

@@ -43,7 +43,8 @@ WORKDIR /home/$DOCKER_USER
 ENV PATH="/home/$DOCKER_USER/.local/bin:${PATH}"
 
 # Install uv for quicker package installations (installed to ~/.local/bin with --user)
-RUN python -m pip install --user uv==0.9.29
+ENV UV_NO_CACHE=1
+RUN python -m pip install --no-cache-dir --user uv==0.9.29
 
 # Create virtual environment with uv
 RUN uv venv /home/$DOCKER_USER/.venv

ML-Frameworks/pytorch-aarch64/examples/executor/model.py

@@ -8,7 +8,6 @@
 import os
 import runpy
 import time
-import sys
 from urllib.parse import urlparse
 import warnings
 import zipfile

ML-Frameworks/pytorch-aarch64/examples/quantized_linear.py

@@ -67,4 +67,4 @@ def forward(self, x):
         model(data)
         runtimes.append(time.time() - t0)
 
-print('Quantized: %.4fms, FP32: %.4fms, Speedup: %.4f' % (np.min(runtimes)*1e3, np.min(fp32_runtimes)*1e3, np.min(fp32_runtimes)/np.min(runtimes)))
+print('Quantized: %.4fms, FP32: %.4fms, Speedup: %.4f' % (np.min(runtimes)*1e3, np.min(fp32_runtimes)*1e3, np.min(fp32_runtimes)/np.min(runtimes)))

ML-Frameworks/pytorch-aarch64/examples/transformers_llm_text_gen.py

@@ -158,8 +158,7 @@ def eval_quantized_output(quantized_model, tokenizer, input_tensor, max_min_toke
 
 
 def main(args):
-    name_string = f"{args.model}"
-    quantized_model_, tokenizer_, config_ = get_quantized_model(args)
+    quantized_model_, tokenizer_, _ = get_quantized_model(args)
     input_tensor = tokenizer_.encode(args.prompt, return_tensors="pt")
     eval_quantized_output(
         quantized_model_, tokenizer_, input_tensor, args.max_new_tokens

ML-Frameworks/pytorch-aarch64/examples/utils/image.py

@@ -247,7 +247,8 @@ def _postprocess_image_for_openimages_detection(
 
         # resizing the box values from the 800x800 image to the original
         # resolution.
-        resize = lambda x, orig: int((x / 800) * orig)
+        def resize(x, orig):
+            return int((x / 800) * orig)
         left = resize(box[0], width)
         top = resize(box[1], height)
         right = resize(box[2], width)

ML-Frameworks/pytorch-aarch64/examples/utils/vision_parser.py

@@ -61,7 +61,7 @@ def parse_arguments():
     else:
         try:
             os.path.isfile(args["image"])
-        except:
+        except Exception as _:
             assert False, "Image not found"
 
     return args

ML-Frameworks/pytorch-aarch64/get-source.sh

@@ -144,4 +144,3 @@ git-shallow-clone https://github.com/pytorch/pytorch.git $PYTORCH_HASH
         )
     fi
 )
-

ML-Frameworks/tensorflow-aarch64/.gitignore

@@ -1,7 +1,7 @@
-# SPDX-FileCopyrightText: Copyright 2025 Arm Limited and affiliates.
+# SPDX-FileCopyrightText: Copyright 2025, 2026 Arm Limited and affiliates.
 #
 # SPDX-License-Identifier: Apache-2.0
 
 tensorflow/
 results/
-*.whl
+*.whl

ML-Frameworks/tensorflow-aarch64/CHANGELOG.md

@@ -29,7 +29,7 @@ where `YY` is the year, and `MM` the month of the increment.
  - Updates TensorFlow hash to 7fd0883713c3c95b4ace0a637de461c731c1db1d from nightly, July 7th
 
 ### Removed
- - Removes TensorFlow [PR #102272](https://github.com/tensorflow/tensorflow/pull/102272), which fixes AArch64 CPUIDInfo init. 
+ - Removes TensorFlow [PR #102272](https://github.com/tensorflow/tensorflow/pull/102272), which fixes AArch64 CPUIDInfo init.
  It was merged upstream on 17 June.
 
 ### Fixed

ML-Frameworks/tensorflow-aarch64/Dockerfile

@@ -38,7 +38,8 @@ WORKDIR /home/$DOCKER_USER
 ENV PATH="/home/$DOCKER_USER/.local/bin:${PATH}"
 
 # Install uv for quicker package installations (installed to ~/.local/bin with --user)
-RUN python -m pip install --user uv==0.9.29
+ENV UV_NO_CACHE=1
+RUN python -m pip install --no-cache-dir --user uv==0.9.29
 
 # Create virtual environment with uv
 RUN uv venv /home/$DOCKER_USER/.venv

ML-Frameworks/tensorflow-aarch64/examples/executor/model.py

@@ -13,7 +13,6 @@
 import zipfile
 
 import numpy as np
-import yaml
 from tqdm import tqdm
 
 import tensorflow as tf

ML-Frameworks/tensorflow-aarch64/examples/utils/vision_parser.py

@@ -89,7 +89,7 @@ def parse_arguments():
     else:
         try:
             os.path.isfile(args["image"])
-        except:
+        except Exception as _:
             assert False, "Image not found"
 
     return args