compare-results

作成者: nvidia

ベースラインと候補の評価計画を確立し、不足している評価を委任し、検証済みの結果を比較し、量子化の実現可能性を判断する。以下の場合に使用する…

npx skills add https://github.com/nvidia/model-optimizer --skill compare-results

Compare Results

Use this to plan and complete a baseline-vs-candidate comparison. The baseline is the reference checkpoint, and the candidate is the checkpoint whose accuracy change is being measured, typically a further quantized version of the baseline.

Workflow

  1. Establish the candidate checkpoint/run and the matching baseline. Infer the baseline from the PTQ source model/checkpoint in the workspace or config used to create the candidate. If it cannot be inferred, ask the user for the baseline checkpoint or an existing baseline invocation/run path.
  2. If a required baseline or candidate evaluation is missing, delegate to the evaluation skill to create, run, and verify it. The companion evaluation config should match benchmark versions, task configs, serving args, token limits, dataset setup, credentials, cluster, and container as closely as possible; change only the model/checkpoint and checkpoint-specific serving or quantization flags.
  3. Fetch the baseline and candidate task list, configs, score artifacts, and logs. If the user provides MLflow runs or invocation IDs, use the accessing-mlflow skill to fetch configs and artifacts.
  4. Confirm each run passed evaluation Step 9, "Verify completed evaluation run", before comparing scores. If not, validate logs, server health, judge/code-execution status, sample accounting, and reasoning parsing before computing deltas.
  5. For each task, use the canonical score field from the matching evaluation skill task recipe, recipes/tasks/<task>.md, under Score Extraction.
  6. Use the evaluation skill's references/run-validation.md to perform the External Baseline Sanity Check. Record each source URL, protocol difference, and task status before applying the candidate-delta gate. A failed baseline blocks a success verdict; correct and rerun it first. If no credible comparable reference exists, label the baseline externally unverified rather than claiming the check passed, then continue using the validated measured baseline.
  7. Compute exact deltas outside the chat context when there are multiple tasks or repeated runs.
  8. Report comparability, external baseline sanity, and quantized-feasibility verdicts before interpreting the delta as model quality. If the user did not provide an acceptance threshold, report feasibility as inconclusive instead of inventing one.

Comparability Checklist

Before treating a baseline-vs-quantized delta as a model quality result, verify the validated runs are comparable:

  1. Prompt text, system prompt, chat template, and rendered messages match.
  2. Task name, benchmark version, dataset split, container, harness, and task fragment match.
  3. Generation settings match, including temperature, top_p, top_k, max tokens, stop strings, chat-template kwargs, reasoning mode/budget, and task-specific overrides.
  4. Reasoning traces are enabled, disabled, parsed, stripped, or ignored consistently between runs.
  5. The number of evaluated and scored samples/repeats matches for each task and split.
  6. Judge-backed or simulator-backed tasks use the same judge/user model, endpoint class, prompt, and scoring config.
  7. The same accuracy metric and score field is used for both runs.
  8. Baseline precision matches the gate. A <1pp vs BF16 gate requires a true full-precision (BF16) baseline. Many models ship natively quantized (e.g. INT4 W4A16 or block-wise FP8) with no BF16 release — a quant-to-quant comparison against the released precision (e.g. INT4 vs NVFP4, as for Kimi-K2.6) is still a valid result; just compare like-for-like, state which precision the baseline is, and apply the gate relative to that baseline rather than to an assumed BF16.

For SciCode, keep num_repeats: 1 to limit sandbox workload. If variance is a concern, run multiple independent matched baseline/candidate pairs instead of increasing repeats within one run.

If any item differs, either rerun with matched settings or label the result as not an apples-to-apples quantization comparison.

These checks compare the baseline and candidate to each other. The external baseline check in the evaluation skill's references/run-validation.md separately tests whether the baseline's absolute score is credible; both guards must be reported.

Report Format

Include:

  • Baseline and candidate identifiers.
  • Per-task metric path, baseline score, candidate score, delta, and stderr if available.
  • Per-task external reference score, source URL, known protocol differences, percentage-point difference, and sanity status (verified, failed, or externally unverified).
  • Comparability status for prompt/template, generation settings, sample counts, reasoning handling, judge/simulator setup, and score field.
  • Comparability verdict: comparable, not comparable, or inconclusive.
  • Quantization feasibility verdict: acceptable, not acceptable, or inconclusive. Never report acceptable when external baseline sanity failed. An externally unverified baseline does not block acceptable; apply the candidate-delta gate and report the missing external corroboration.

nvidiaのその他のスキル

compileiq-debug
nvidia
何かがおかしいときに使用:Search()がハングする、すべての評価がINVALID_SCOREを返す、スコアが改善しない、すべての設定が同じ数値を返す、ptxasエラー…
create-github-pr
nvidia
gh CLIを使用してGitHubのプルリクエストを作成します。ユーザーが新しいPRを作成したい、コードをレビューに提出したい、またはプルリクエストを開きたい場合に使用します。トリガーキーワード -…
nemoclaw-maintainer-cross-issue-sweep
nvidia
他のオープンなIssueをスキャンし、特定のPRが修正する可能性があるものや、誤って壊す可能性があるものを見つけます。隣接修正の機会や矛盾リスクをfile:line…と共に出力します。
fhir-basics
nvidia
エージェントにFHIR R4 APIの動作方法、利用可能なリソース、検索パラメータを使ったクエリ方法、およびすべてのレスポンス形式を正しく解析する方法を教えます…
compileiq-validate-result
nvidia
検索が完了した後、かつスピードアップの申請やACFの発送の前に使用します。dump_results CSVを読み込み、トップK候補(単一目的)を抽出します…
changelog-audit
nvidia
リリース前にWarp CHANGELOG.mdを監査:失われたエントリを復元、ユーザー影響で並べ替え、エントリの文言を洗練、行折り返し、および(リリースブランチモードで)比較をバンプ…
maintain-dynamic-plugins
nvidia
NeMo Relayの動的プラグインローダー、マニフェスト、RustネイティブSDK、gRPCワーカープロトコル、PythonワーカーSDK、ドキュメント、テスト、およびリリースワークフローのカバレッジを維持する
dgx-diagnose
nvidia
一般的なDGX Station GB300の問題(CUDAクラッシュ、誤ったGPUターゲット、vLLM/SGLangコンテナのバグ、MIG状態の問題、NVLink/Fabric Managerエラーなど)を診断します。