Integrates new tabular ML models into the TabArena benchmark as AutoGluon wrappers.
npx skills add https://github.com/autogluon/tabarena --skill add-model-autogluon
Please help me install this Agent Skill. Skill: add-model Source: https://github.com/autogluon/tabarena/tree/main/.claude/skills/add-model Command: npx skills add https://github.com/autogluon/tabarena --skill add-model-autogluon
packages/tabarena/src/tabarena/models/<ModelKey>/ with __init__.py, model.py, hpo.py, and info.py, auto-discovered by the model registry with no per-model test file needed. - Base-class and pattern guidance: Selects AbstractTorchModel vs AbstractModel based on framework (torch, JAX/Flax, sklearn-like), enforces the _fit contract (validation split, time limit, CPU/GPU budgets), and covers warm-up declarations and shared pretrained weights for foundation models. - Registry and packaging edits: Updates the lazy-import map, pyproject.toml extras (benchmark/extended/all unions), and the leaderboard model-family mapping, with a drift checker to keep extras in sync. - Use Case: You want to benchmark a new foundation model like TabPFN-2.6 in TabArena. Provide the model name, pip package, and docs URL, and the Skill generates the wrapper, search space, metadata, and packaging edits following in-repo reference implementations. ## Quick Start Add the TabStar model to the TabArena benchmark using the tabstar pip package and its GitHub documentation URL.High-intent search queries and answers about installing and using this skill.
Create a folder under packages/tabarena/src/tabarena/models/<ModelKey>/ containing model.py (the AutoGluon wrapper), hpo.py (search-space generator), info.py (ModelInfo and MethodMetadata), and an __init__.py. The registry auto-discovers it, and test_all_models.py fit-tests it automatically.
Use AbstractTorchModel only for torch-based models, since it provides get_device and _set_device device management. Non-torch GPU models like JAX/Flax libraries and CPU sklearn-like models inherit AbstractModel and add GPU resource attributes without implementing device methods.
No per-model test file is needed. tests/tabarena/models/test_all_models.py is parametrized over the model registry and fits the new model automatically once its info.py is discoverable, skipping on missing optional dependencies or unavailable GPUs.
Declare a shared_weights ClassVar naming the library's loader call and the inputs that decide which network is built. AutoGluon's AbstractTorchModel memoizes the network per process so bagged folds and refits reuse one build instead of reloading per fold.
The model's config_type was not added to get_model_family in website_format.py. Add the uppercased config_type to the correct family list (Tree-based, Foundation Model, Neural Network) so prefix matching assigns the proper leaderboard Type.
Use add-system when the integration is a self-contained ML system that does its own preprocessing, validation, HPO, or ensembling, such as AutoML frameworks, multi-model stacks, or agent pipelines. Single models fit by TabArena's shared harness use add-model.
minhquan2904Scan codebases for factory patterns and generate knowledge_factory.md files.
surus-latMerge local benchmark results into the HuggingFace LatamBoard leaderboard dataset.
opensage-agentOrchestrate multi-agent systems with ensemble, pool, and retry patterns.
naviktGuide Nais manifest creation and Kubernetes troubleshooting for deployment issues.