simpo-training
关于
SimPO is a reference-free LLM alignment method that serves as a simpler, more efficient alternative to DPO. It eliminates the need for a reference model while achieving better performance, such as a +6.4 point improvement on AlpacaEval 2.0. Use it for preference alignment when you want faster, less complex training than DPO or PPO.
快速安装
Claude Code
推荐npx skills add davila7/claude-code-templates -a claude-code/plugin add https://github.com/davila7/claude-code-templatesgit clone https://github.com/davila7/claude-code-templates.git ~/.claude/skills/simpo-training在 Claude Code 中复制并粘贴此命令以安装该技能
GitHub 仓库
Frequently asked questions
What is the simpo-training skill?
simpo-training is a Claude Skill by davila7. Skills package instructions and resources that Claude loads on demand, so Claude can perform simpo-training-related tasks without extra prompting.
How do I install simpo-training?
Use the install commands on this page: add simpo-training to Claude Code as a plugin, or clone its repository into your skills directory, then restart Claude so it picks up the skill.
What category does simpo-training belong to?
simpo-training is in the Other category, tagged Post-Training, SimPO, Preference Optimization, Alignment, DPO Alternative and Reference-Free.
Is simpo-training free to use?
Yes. simpo-training is listed on AIMCP and free to install. It runs inside Claude, so no separate service account is required to use the skill itself.
相关推荐技能
LlamaGuard是Meta推出的7-8B参数内容审核模型，专门用于过滤LLM的输入和输出内容。它能检测六大安全风险类别（暴力/仇恨、性内容、武器、违禁品、自残、犯罪计划），准确率达94-95%。开发者可通过HuggingFace、vLLM或Sagemaker快速部署，并能与NeMo Guardrails集成实现自动化安全防护。
这个Claude Skill帮助开发者优化云成本，通过资源调整、标记策略和预留实例来降低AWS、Azure和GCP的开支。它适用于减少云支出、分析基础设施成本或实施成本治理策略的场景。关键功能包括提供成本可视化、资源规模调整指导和定价模型优化建议。
该Skill为开发者提供体育博彩数据分析工具，可分析盘口、大小球和特殊投注，识别价值投注机会。它整合历史数据和情景统计，生成包含时间戳的结构化Markdown报告。适用于需要快速获取博彩市场洞察的娱乐或教育类应用开发。
这个Skill使用bitsandbytes库量化大语言模型，能在GPU内存有限时通过8位或4位量化减少50-75%内存占用，同时保持精度损失最小。它支持INT8、NF4、FP4等多种量化格式，可与HuggingFace Transformers无缝集成，适用于需要部署更大模型或加速推理的场景。还提供QLoRA训练和8位优化器支持，让开发者能轻松实现高效模型压缩。