TorchJD
Multi-task PyTorch training with conflict-free gradient aggregation
All Skills in This Repository (4)
Pure Emerald Level Indicators
SimplexLabcheck-release
Verify TorchJD releases via docs, PyPI installation, and API smoke tests.
SimplexLabresearch-method
Research scientific literature and map optimization methods to TorchJD integration requirements.
SimplexLabimplement-method
Implement TorchJD methods from research into repository artifacts.
SimplexLabprepare-release
Verify changelog, README, and version in pyproject.toml for TorchJD releases.
Frequently Asked Questions
FAQPage SchemaHow to install TorchJD?▼
Run `npx skills add SimplexLab/TorchJD --all -g -y` in your terminal to install all skills in this suite globally.
What problem does TorchJD solve?▼
It trains neural networks with multiple losses by aggregating conflicting gradients into a single update direction, which standard loss summation often fails to do.
How to add a new aggregation method to TorchJD?▼
Use the research-method skill to analyze the paper, then the implement-method skill to generate the class, tests, docs, and changelog following project conventions.
Does TorchJD work with standard PyTorch training loops?▼
Yes. You only replace the loss summation and backward call with a scalarizer or the mtl_backward and jac_to_grad functions.
How to verify a TorchJD release?▼
The check-release skill validates the docs site, installs the package from PyPI, and smoke-tests all newly added classes automatically.
Related Repositories in Software Engineering
View All in Software Engineering→
openclawopenclaw
Run a personal AI assistant across your devices and chat apps
obrasuperpowers
Gives coding agents a disciplined workflow from idea to merged code
facebookreact
AI agent skills for building, testing, and porting React core