菜单
Synergistic verifier-generator training loop enabling LLMs to identify logical issues in mathematical proofs without reference solutions, improving reasoning rigor through meta-verification. Apply when you need to scale mathematical reasoning without hand-labeled proof annotations.
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。
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/ADu2021/skillXiv --skill deepseek-math-v2-self-verifiable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。
Convert arXiv papers into ready-to-use agent skills using category-aware extraction. First classifies the paper into one or more of 11 research categories, then applies a specialized extraction pipeline for each category — because different types of papers produce different types of usable knowledge. A single paper can yield multiple skills if it spans categories. Use this skill whenever the user wants to turn a paper into a skill, extract practical techniques from research, build a skill library from papers, convert arXiv papers into reusable agent instructions, or batch-process multiple papers into skills. Also trigger when someone asks about extracting actionable knowledge from papers, making research practical for LLM agents, or systematically converting academic contributions into structured agent capabilities.
Establish regret bounds for behavior cloning with discretized actions combining statistical error and quantization error terms. Prove smoothness requirements for safe quantizer design, show that learning-based quantizers fail these requirements, and propose model-based augmentation to reduce error dependence from H² to H.
Dynamically allocate LoRA ranks per-layer during fine-tuning instead of using fixed uniform ranks. Learn optimal rank for each layer and subject via variational framework with discretized exponential distribution, reducing memory footprint while maintaining fidelity and text-alignment.
基于 SOC 职业分类
正在显示 SKILL.md
| name | deepseek-math-v2-self-verifiable |
| title | DeepSeekMath-V2: Self-Verifiable Mathematical Reasoning Through Iterative Training |
| version | 0.0.2 |
| engine | skillxiv-v0.0.2-claude-opus-4.6 |
| license | MIT |
| url | https://arxiv.org/abs/2511.22570 |
| keywords | ["mathematical-reasoning","proof-verification","self-verification","iterative-training","llm-alignment"] |
| description | Synergistic verifier-generator training loop enabling LLMs to identify logical issues in mathematical proofs without reference solutions, improving reasoning rigor through meta-verification. Apply when you need to scale mathematical reasoning without hand-labeled proof annotations. |
DeepSeekMath-V2 introduces a self-verifiable mathematical reasoning system through an iterative training cycle between proof verification and generation components. Instead of using final-answer rewards, the approach directly addresses proof rigor by training an LLM-based verifier that identifies logical issues and a generator that performs self-verification, creating synergistic improvement cycles.
The system operates through three coupled components:
Verifier Training: Train an LLM to identify logical issues in proofs using meta-verification—where the verifier itself must identify which rubrics it should evaluate against. This eliminates the need for reference solutions or ground-truth proof steps.
Generator Self-Verification: Train the proof generator to analyze its own work using the same rubrics as the external verifier, enabling deliberate self-refinement during generation rather than blind trial-and-error.
Verification-Scaled Labeling: Use the verifier to automatically label challenging proofs, creating training data that progressively improves both components in a reinforcement cycle.
Meta-verification: The verifier outputs both a proof evaluation and the reasoning criteria it applied. This self-aware approach ensures the verifier produces faithful analyses rather than hallucinated critiques.
Synergistic scaling: Allocate compute to verification as a first-class post-training component, not just as an evaluation metric. Scaling verification compute directly improves the verifier's ability to label and improve the generator.
Iterative refinement: Cycle between: (1) generator creates proofs with self-verification, (2) external verifier identifies remaining issues, (3) both components retrain on improved data, (4) repeat.
