keras-tuner
Routes KerasTuner workflows for defining hyperparameter spaces, running search algorithms, tuning image and scikit-learn models, and coordinating distributed chief/worker search state.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- keras-tuner
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Routes KerasTuner workflows for defining hyperparameter spaces, running search algorithms, tuning image and scikit-learn models, and coordinating distributed chief/worker search state.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/api-reference.md6.6 KB
- 参考文档references/repo-provenance.md1.8 KB
- 参考文档references/repo-routing-metadata.json305 B
- 参考文档references/troubleshooting.md4.3 KB
- 参考文档references/workflows.md3.6 KB
- 脚本scripts/check_env.py2.6 KB
03 SKILL.md 正文
KerasTuner
Use this skill when the task mentions keras_tuner, KerasTuner, HyperParameters, HyperModel, Oracle, Tuner, RandomSearch, GridSearch, Hyperband, BayesianOptimization, SklearnTuner, HyperResNet, HyperXception, HyperEfficientNet, HyperImageAugment, or distributed KERASTUNER_* search state.
What this skill covers
- Defining search spaces with
HyperParametersandHyperModel. - Running the built-in search algorithms and reading back best trials.
- Tuning scikit-learn estimators with
SklearnTuner. - Using the built-in image hypermodels and augmentation search helpers.
- Debugging distributed chief/worker oracle coordination.
Install and quick check
For a normal CPU-backed KerasTuner install:
python -m pip install "keras-tuner[tensorflow-cpu]" tensorboard
For Bayesian optimization or scikit-learn tuning, add the optional backend package set:
python -m pip install "keras-tuner[tensorflow-cpu,bayesian]" tensorboard
If you are using a GPU TensorFlow environment, swap tensorflow-cpu for tensorflow.
For editable development from a source checkout, run this from the checkout
root that contains setup.py; the generated skill directory is documentation
and is not itself an installable checkout:
python -m pip install -e ".[tensorflow-cpu,bayesian]" tensorboard
Minimal check:
python -c "import keras_tuner; from keras_tuner.backend import config; print(keras_tuner.__version__, config.backend())"
Run scripts/check_env.py when you want a slightly richer import and backend sanity check.
Route map
- Read
sub-skills/tuning/SKILL.mdfor generic hyperparameter search, customHyperModelclasses,Oracle/Tunersubclassing, trial inspection, and reload/resume workflows. - Read
sub-skills/image-hypermodels/SKILL.mdforHyperResNet,HyperXception,HyperEfficientNet, andHyperImageAugment. - Read
sub-skills/sklearn-tuning/SKILL.mdforSklearnTuner, estimator factories, cross-validation, sample weights, and pickle-backed model saving. - Read
sub-skills/distributed-tuning/SKILL.mdforKERASTUNER_ORACLE_IP,KERASTUNER_ORACLE_PORT,KERASTUNER_TUNER_ID, and local chief/worker coordination.
What to read next
references/api-reference.mdfor verified signatures and class responsibilities.references/workflows.mdfor end-to-end tuning patterns.references/troubleshooting.mdfor install/import, optional dependency, and search failures.references/repo-provenance.mdbefore deciding whether this skill matches the current checkout.
Operational notes
- This package is Python-first; there is no separate CLI to route here.
- Use
keras_tuner.backend.config.backend()to check the active backend, notkeras_tuner.backend.backend. - Bayesian optimization and scikit-learn tuning need the optional
bayesiandependency set. - In this 1.4.8 TensorFlow path, Keras model trials also import TensorBoard's HParams API; install
tensorboardeven when you do not pass a TensorBoard callback. - The built-in image models may be slow to build;
HyperEfficientNetcan also trigger a first-run Keras Applications weight download. - Distributed tuning needs the chief/worker environment variables to be set together.
Unless a command says otherwise, run the shell snippets from this skill's root directory (the directory containing this
SKILL.md); from another cwd, use the corresponding absolute path.
Fast path
If you only need to confirm the install, backend, and public exports, run:
python scripts/check_env.py
If you need a workflow-specific smoke check, run the helper from the skill
root, using its full bundle-relative path, such as
python sub-skills/tuning/scripts/smoke_search.py.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).