benchmark-cli
Use LimiX benchmark-style classification and regression CLI commands with safe local dataset and checkpoint handling.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- benchmark-cli
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use LimiX benchmark-style classification and regression CLI commands with safe local dataset and checkpoint handling.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/cli-reference.md8.6 KB
- 参考文档references/dataset-layout.md4.7 KB
- 参考文档references/troubleshooting.md7.3 KB
- 脚本scripts/validate_dataset_layout.py16.0 KB
03 SKILL.md 正文
benchmark-cli
Use this sub-skill when a task asks for LimiX benchmark-style CLI inference over one or more local tabular dataset folders, result CSV collection, or dataset-layout validation before running the classification/regression benchmark scripts.
Start here
- Validate the dataset root before model inference:
python sub-skills/benchmark-cli/scripts/validate_dataset_layout.py DATASET_ROOT --task classification python sub-skills/benchmark-cli/scripts/validate_dataset_layout.py DATASET_ROOT --task regression - Prefer local inputs: pass both
--data_dirand--model_path. Omitting either can trigger network downloads into./cache. - Choose an explicit config JSON path instead of relying on missing default config filenames.
- Treat full checkpoint CLI inference as GPU/CUDA-sensitive: the benchmark scripts guard on CUDA availability before they process workloads, and full LimiX checkpoint inference needs a local LimiX checkpoint. Do not claim a benchmark CLI was run unless it actually completed.
Reference map
references/cli-reference.md: flag meanings, safe local command templates, DDP launch shapes, output files, and metrics.references/dataset-layout.md: expected dataset-folder schema, target-column placement, train/test behavior, dtype/category handling, and validator usage.references/troubleshooting.md: GPU guard, auto-download/network failures, class-count and row-count skips, data-shape errors, DDP notes, and result-path confusion.
Routing boundaries
- For direct
LimiXPredictorPython API usage, route to../predictor-inference/SKILL.md. - For inference config schema details, config generation, or preprocessing-pipeline edits, route to
../configuration-preprocessing/SKILL.md. - For retrieval tuning/search-space semantics beyond the CLI flag surface, route to
../retrieval-optimization/SKILL.md. - Do not run heavyweight benchmark CLIs as a validation shortcut. Use the bundled dataset validator and provide command recipes unless the user explicitly authorizes full inference with local data, a local checkpoint, and a ready CUDA/GPU environment.
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).