benchmark-evaluation
Guide Otter config-driven benchmark evaluation, registries, GPT-judged requirements, logging, cache/model paths, and public suite caveats.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- benchmark-evaluation
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Guide Otter config-driven benchmark evaluation, registries, GPT-judged requirements, logging, cache/model paths, and public suite caveats.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/benchmark-configs.md5.0 KB
- 参考文档references/model-and-dataset-registry.md6.7 KB
- 参考文档references/public-suite.md5.5 KB
- 参考文档references/troubleshooting.md5.2 KB
- 脚本scripts/validate_benchmark_config.py17.1 KB
03 SKILL.md 正文
Otter benchmark-evaluation sub-skill
Use this sub-skill when the task is to plan, validate, or run Otter benchmark evaluation with pipeline.benchmarks.evaluate, benchmark YAMLs, supported model/dataset registry names, GPT-judged benchmark credentials, output logs, cache directories, or the OpenFlamingo-style public dataset suite.
Fast route
- Pick the evaluator path:
- Config-driven Otter evaluator: benchmark-configs.
- Supported
models[].nameanddatasets[].namekeys: model-and-dataset-registry. - Public COCO/VQA/Flickr/TextVQA/VizWiz/Hateful-Memes/ImageNet suite: public-suite.
- Validate the benchmark YAML before launching model or dataset downloads:
python scripts/validate_benchmark_config.py benchmark.yaml
- Launch with the correct flag name after validation:
python -m pipeline.benchmarks.evaluate --config benchmark.yaml
The source docs contain a typo, --confg; use --config or -c.
- Before expensive or networked runs, check troubleshooting for GPT API keys, Hugging Face dataset downloads, missing local model repos, cache placement, and expected skip reasons.
Route elsewhere
- Training and finetuning launch construction: training.
- MIMIC-IT, Syphus, Convert-It, and large data conversion: data-preparation.
- Controller/worker/Gradio/API serving: serving.
- Prompt/media tensor details for ad hoc generation outside benchmarks: model-inference.
Safety boundary
Benchmark evaluation can load large models, download benchmark datasets, write result files, and call paid GPT APIs. This skill supports static planning and config validation; only run evaluation when the user has supplied model paths or approved downloads, dataset/cache/output locations, required credentials, and GPU/runtime budget.
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).