configuration-preprocessing
Inspect, choose, generate, and debug LimiX inference configs and preprocessing transforms.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- configuration-preprocessing
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Inspect, choose, generate, and debug LimiX inference configs and preprocessing transforms.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/config-catalog.md8.5 KB
- 参考文档references/preprocessing-reference.md8.3 KB
- 参考文档references/troubleshooting.md8.2 KB
- 脚本scripts/generate_noretrieval_config.py8.5 KB
- 脚本scripts/inspect_config.py14.5 KB
03 SKILL.md 正文
LimiX configuration and preprocessing router
Use this sub-skill when the task is to choose, validate, generate, or debug a LimiX inference configuration JSON or the preprocessing blocks that a config enables.
Do not use this sub-skill to run the model end to end. Full checkpoint inference needs a local LimiX checkpoint and may require CUDA/GPU; this sub-skill only covers configuration and preprocessing decisions.
Route by intent
- Need to call
LimiXPredictor, preparex_train/y_train/x_test, inspect outputs, or use missing-value imputation? Load../predictor-inference/SKILL.md. - Need to run the benchmark-style classification/regression CLI over dataset directories? Load
../benchmark-cli/SKILL.md. - Need retrieval search, Optuna tuning, attention-map semantics, or retrieval-parameter trade-offs? Load
../retrieval-optimization/SKILL.md. - Need to inspect whether a JSON config is valid, CPU-safe, retrieval-enabled, MVI-compatible, or likely to fail in preprocessing? Continue here.
Operating checklist
- Pick the nearest catalog config from
references/config-catalog.mdby task, model size, retrieval use, and MVI need. - For CPU or low-memory work, prefer non-retrieval configs; LimiX rejects retrieval configs on CPU and retrieval inference can be GPU-memory heavy.
- Validate user-supplied JSON with
scripts/inspect_config.pybefore using it in a predictor or benchmark command. - If no config file exists, generate a safe no-retrieval baseline with
scripts/generate_noretrieval_config.py, then inspect it. - When a preprocessing failure mentions constant/all-NaN features, category encoding, power transforms, one-hot size, KDI, Hyperopt, or malformed config keys, use
references/troubleshooting.md.
Bundled references and scripts
references/config-catalog.md: observed config matrix, config-list schema, helper-generated configs, and CPU/retrieval caveats.references/preprocessing-reference.md: transform order, class behavior, key options, data implications, and failure modes.references/troubleshooting.md: actionable fixes for malformed configs and preprocessing/runtime errors.scripts/inspect_config.py: standalone JSON validator and config summarizer.scripts/generate_noretrieval_config.py: standalone no-retrieval config generator.
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).