v0.10.0

[AlexeyKozhevin]

This pull request introduces significant improvements to the release automation workflow and adds enhanced support for Apple's Metal (MPS) backend in PyTorch models. It also bumps the project version to 0.10.0. Below are the most important changes grouped by theme:

Release Automation Overhaul:

• Replaces the old release workflows (release.yml and release-check.yml) with a new, modular set of GitHub Actions:
  • .github/workflows/release-on-branch-create.yml: Automatically validates and prepares release branches, updates version numbers, and creates draft pull requests for new releases.
  • .github/workflows/release-on-merge.yml: Handles release tagging, PyPI/TestPyPI publishing, next release branch creation, and hotfix propagation after merging release PRs into master.
• Removes legacy release workflows to avoid conflicts and streamline the process. [1] [2]

PyTorch MPS (Apple Silicon) Support:

• Updates device parsing and selection logic in batchflow/models/torch/base.py to recognize and use the MPS backend when available, in addition to CUDA and CPU.
• Ensures device lists and device-specific operations (autocast, GradScaler, etc.) are compatible with MPS, not just CUDA. [1] [2] [3] [4] [5] [6]
• Adjusts AMP (automatic mixed precision) and backend benchmarking logic to be device-agnostic.

Version Bump:

• Increments the project version in pyproject.toml from 0.9.4 to 0.10.0 to reflect these breaking and feature changes.

[Copilot]

Pull request overview

This pull request introduces version 0.10.0 with two major feature sets: a complete overhaul of the release automation workflow and enhanced support for Apple's Metal Performance Shaders (MPS) backend in PyTorch models.

Changes:

• Replaced legacy GitHub Actions workflows (release.yml and release-check.yml) with modular release automation supporting branch-based versioning, automated PyPI publishing, and hotfix propagation
• Added MPS device detection and selection in PyTorch model device parsing to enable Apple Silicon GPU acceleration
• Updated PyTorch operations (autocast, GradScaler, benchmark configuration) to be device-type aware rather than CUDA-specific
• Bumped project version from 0.9.4 to 0.10.0 in pyproject.toml and uv.lock

Reviewed changes

Copilot reviewed 6 out of 7 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
.github/workflows/release-on-branch-create.yml	New workflow to automatically prepare release branches with version updates and draft PRs
.github/workflows/release-on-merge.yml	New workflow handling release tagging, PyPI publishing, next release branch creation, and hotfix propagation
.github/workflows/release.yml	Removed legacy release workflow to avoid conflicts
.github/workflows/release-check.yml	Removed legacy release check workflow to avoid conflicts
batchflow/models/torch/base.py	Added MPS device support in device parsing, updated autocast and GradScaler to use device type instead of hardcoded 'cuda', adjusted AMP disabling logic and cudnn benchmark configuration
pyproject.toml	Version bumped from 0.9.4 to 0.10.0
uv.lock	Updated to reflect new version 0.10.0 and regenerated lock file

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.