tasks-and-benchmarks
Discover, filter, inspect, and reason about MTEB tasks and predefined benchmarks before evaluation.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- tasks-and-benchmarks
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Discover, filter, inspect, and reason about MTEB tasks and predefined benchmarks before evaluation.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/metadata-and-filters.md5.4 KB
- 参考文档references/task-benchmark-reference.md4.6 KB
- 参考文档references/troubleshooting.md5.1 KB
- 脚本scripts/inspect_task_selection.py10.5 KB
03 SKILL.md 正文
MTEB Tasks and Benchmarks
Use this sub-skill when the user needs to choose MTEB tasks or benchmarks, inspect task metadata, filter by language/script/domain/task type/category/modality, understand beta/private/superseded defaults, or compare benchmark task lists.
Route First
- For running selected tasks or benchmarks with a model, use
../evaluation-workflows/SKILL.mdafter selection is complete. - For shell workflows around
mteb available-tasksormteb available-benchmarks, use../cli-and-automation/SKILL.md. - For implementing or registering new tasks, use
../contributing-to-mteb/SKILL.md.
Core Workflows
- Select named tasks with
mteb.get_tasks(tasks=[...])or one task withmteb.get_task(name, languages=..., script=..., eval_splits=..., hf_subsets=...); named-task selection ignores broaddomains,task_types, andcategoriesfilters, so applymteb.filter_tasks(...)to a returned list if both name and metadata filtering are required. - Discover tasks with
mteb.get_tasks(languages=..., script=..., domains=..., task_types=..., categories=..., modalities=..., eval_splits=...); filters are combined as an AND operation, with OR semantics inside each list unless an exclusive flag is documented. - Select benchmarks with
mteb.get_benchmark("MTEB(eng, v2)")or enumerate them withmteb.get_benchmarks(names=..., display_on_leaderboard=...); aBenchmarkis iterable and can be passed astasks=to evaluation. - Filter benchmark tasks with
mteb.filter_tasks(benchmark, task_types=[...], languages=[...], modalities=[...]), then pass the filtered list to evaluation. - Inspect selections safely with
scripts/inspect_task_selection.py; it imports MTEB and prints task/benchmark summaries without downloading datasets or running models.
References
references/task-benchmark-reference.md- API patterns for task and benchmark discovery, selection, and benchmark filtering.references/metadata-and-filters.md- metadata fields, filter semantics, language/script distinctions, and default exclusion behavior.references/troubleshooting.md- diagnosis for import issues, optional dependencies, private/beta/superseded surprises, task-name typos, and CLI/API misuse.
Quick Validation
- Confirm package import with
python -c "import mteb; print(mteb.__version__)". - Preview a task filter with
python scripts/inspect_task_selection.py --task-types Retrieval --languages eng --format markdown --limit 20. - Preview a benchmark with
python scripts/inspect_task_selection.py --benchmark "MTEB(eng, v2)" --task-types Retrieval --format json --limit 10. - If selection looks correct, hand the task objects, task names, or benchmark to
evaluation-workflowsfor actualmteb.evaluate(...)usage.
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).