菜单
Compare an AReno branch, model, checkpoint, algorithm, scheduler, or kernel against a baseline. Use for regression validation, logprob or metric comparison, checkpoint round trips, and correctness gates. Keep performance reporting separate.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
基于 SOC 职业分类
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/inclusionAI/AReno --skill areno-validate-correctness命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
Autonomously research, implement, train, evaluate, visualize, and submit a genuinely new AReno RLVR mini-game whose mechanically generated samples are independent state-to-single-output tasks. Use when the user requests a complete new game demo with real reward improvement, not for multi-turn agent workflows, SFT-only examples, or design-only proposals.
Add or debug an AReno model family, including config conversion, module construction, checkpoint load/save, text or multimodal inference, training backward, tensor parallelism, CUDA graph decode, and checkpoint round trips. Use only when model-specific implementation or compatibility changes are required.
Add or modify an AReno algorithm, trainer, loss, advantage calculation, role model, or algorithm-specific configuration. Use for framework-level SFT, DPO, GSPO, GRPO, PPO, or new optimization method development. Do not use merely to run an existing algorithm.
| name | areno-validate-correctness |
| description | Compare an AReno branch, model, checkpoint, algorithm, scheduler, or kernel against a baseline. Use for regression validation, logprob or metric comparison, checkpoint round trips, and correctness gates. Keep performance reporting separate. |
Create an isolated baseline worktree. Keep checkpoint, input, seed, algorithm, topology, dtype, and bounded step count identical.
python .agents/skills/areno-validate-correctness/scripts/compare_metrics.py BASE_DIR CANDIDATE_DIR --metric train/loss
python .agents/skills/areno-validate-correctness/scripts/compare_arrays.py base.json candidate.json --atol 1e-5 --rtol 1e-4
python .agents/skills/areno-model-adaptation/scripts/inspect_checkpoint.py /path/to/checkpoint
Read references/validation-matrix.md. Select invariants based on ownership; final loss alone is insufficient for model, algorithm, scheduler, or kernel changes.
Completion requires reproducible commands, artifacts, comparison output, and a nonzero exit on failed invariants.