AGENT SKILL

result-logging

Understand, validate, summarize, compare, and safely route lm-evaluation-harness result artifacts, sample logs, cache files, and optional external logging integrations.

License 许可
MIT
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
result-logging
license
MIT
metadata
{ "disco-role": "operating" }
description
Understand, validate, summarize, compare, and safely route lm-evaluation-harness result artifacts, sample logs, cache files, and optional external logging integrations.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

Result Logging

Use this sub-skill when the user needs help interpreting saved lm-eval outputs, validating result JSON, comparing finished runs, preparing logging arguments, or converting local result files into tables.

Route Here For

  • Reading aggregated results_*.json files and optional per-task samples_*.jsonl logs.
  • Explaining result schema fields such as results, groups, configs, versions, n-shot, higher_is_better, n-samples, samples, config, task_hashes, tokenizer metadata, and environment metadata.
  • Debugging why metrics, sample logs, task hashes, stderr values, cache files, or service uploads are missing.
  • Preparing --output_path, --log_samples, --predict_only, --wandb_args, --trackio_args, and --hf_hub_log_args without exposing credentials.
  • Summarizing or comparing already-saved local JSON results with bundled scripts.
  • Planning Zeno visualization prerequisites without assuming a network service or API key is available.

Do Not Use For

  • Constructing and running new evaluations; route to ../evaluation-runs/.
  • Choosing or configuring model backends; route to ../model-backends/.
  • Defining task metrics or YAML task behavior; route to ../task-authoring/.
  • Running network uploads to W&B, Trackio Spaces, Hugging Face Hub, or Zeno unless the user explicitly asks and provides safe credential handling.

Fast Paths

  • Result schema: read references/result-schema.md.
  • Logging workflows, table generation, result comparison, service arguments, cache notes, and Zeno prerequisites: read references/result-workflows.md.
  • Failure diagnosis: read references/troubleshooting.md.
  • Local table from one or more result JSON files:
    python scripts/summarize_results.py path/to/results.json
    
  • Local metric comparison between two saved runs:
    python scripts/compare_results.py run_a.json run_b.json --metric acc
    

Safety Defaults

  • Treat result files as local artifacts; do not publish, upload, or create repos unless the user explicitly asks.
  • Never print tokens from HF_TOKEN, W&B, Trackio, Zeno, or config files.
  • Prefer local JSON validation and summaries before diagnosing backend or task behavior.
  • If predict_only is used, expect predictions/sample logs but no aggregate metrics.
  • If log_samples is requested, require an output_path; sample logs are not useful if they are never written.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

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).

★ 351