Llama（大型语言模型）
v1.2.0本地LLM性能优化的完整方法论。发现“上下文甜蜜点” - 上下文减少6%，速度提高75%，且ZERO质量损失。
0· 23·0 当前·0 累计
by 下载技能包
最后更新
2026/4/26
安全扫描
OpenClaw
安全
medium confidence该技能是一份针对优化llama.cpp/llama-server运行时参数的指南；其要求和指令与该目的相一致，但它包含一个潜在风险的推荐，即将服务器绑定到0.0.0.0，用户应该谨慎对待。
评估建议
This guide appears to be a legitimate, self-contained methodology for tuning llama.cpp/llama-server. Before using it: 1) Do not run example server commands that bind to 0.0.0.0 on production or untrusted networks — prefer localhost or restrict access with a firewall / reverse proxy / auth. 2) mlock and very large ctx-size may require root privileges and can exhaust system resources; test on controlled hardware. 3) Use synthetic or non-sensitive prompts when benchmarking to avoid unintentionally ...详细分析 ▾
✓ 用途与能力
Name and description claim a methodology for local LLM performance optimization; SKILL.md contains step‑by‑step benchmarking instructions, parameter lists, and example launch commands. All requested actions (run tests, adjust ctx-size, batch size, flash-attn, kv quantization, etc.) are relevant to the stated purpose.
ℹ 指令范围
Instructions are detailed and stay within deployment and benchmarking scope. Notable caution: example launch commands include --host 0.0.0.0 and a public port (8080), which would expose the model server to the network — this is a security-sensitive deployment choice but is related to running llama-server. The doc also recommends mlock and wide ctx-size testing (resource intensive). The guide does not instruct reading unrelated files, exfiltrating data, or contacting external endpoints.
✓ 安装机制
Instruction-only skill with no install spec and no code files — nothing is downloaded or written by the skill itself. This is the lowest-risk install posture.
✓ 凭证需求
No environment variables, credentials, or config paths are requested. The demands in the guide (GPU, VRAM, CPU binding, large context sizes) are proportional to a performance‑tuning methodology and do not request extraneous secrets or unrelated service credentials.
✓ 持久化与权限
Skill is not always-enabled and does not request persistent privileges or attempt to modify other skills or system settings. Autonomous invocation is allowed by platform default but the skill itself doesn't request elevated or persistent agent-level privileges.
安全有层次，运行前请审查代码。
运行时依赖
无特殊依赖
版本
latestv1.2.02026/4/26
llama-params-optimizer v1.2.0 - 添加了全面的中英双语文档，以提高全球可访问性。 - 改进了描述和关键词，强调了加速（上下文“最佳点”）和通用适用性。 -澄清和简化了步骤式优化方法，以便于采用。 - 突出了真实案例结果（例如，减少6%的上下文可以实现75%更快的速度，零质量损失）。 - 更新了违反直觉的发现和缓解提示，使指南对于多样化的硬件和模型更强健。
● 无害
安装命令
点击复制官方npx clawhub@latest install llama-params-optimizer
镜像加速npx clawhub@latest install llama-params-optimizer --registry https://cn.longxiaskill.com 镜像可用
本土化适配说明
Llama（大型语言模型） 安装说明： 安装命令：npx clawhub@latest install llama-params-optimizer