一个面向工程的本地LLM系统知识库,包含测量结果、条件化报告和死胡同墓碑,设计为工程师或AI代理可直接复用的机制。

Stars

11

7 天增长

暂无数据

Fork 数

0

开放 Issue

0

开源协议

MIT

最近更新

2026-07-28

AI 仓库情报摘要
FR-AI / ANALYSIS

为什么值得关注

它提供可复现的、附带条件的结果,并明确记录了死胡同,同时针对人类工程师和AI代理使用了结构化的Obsidian仓库和可选的MCP服务器。

适合谁使用

  • 本地LLM工程师
  • AI代理开发者
  • Obsidian高级用户
  • 研究投机解码、量化、MoE训练和激活操控的研究人员

典型使用场景

  • 作为子文件夹集成到Obsidian仓库中用于个人知识管理
  • 作为只读知识库供AI代理回答问题或执行任务
  • 探索关于服务、投机解码、量化、MoE训练和激活操控的精选笔记
  • 运行本地MCP服务器以编程方式搜索和检索节点

项目优势

  • 每个重要数字都附带条件,确保可复现性
  • 死胡同作为墓碑保留,为未来研究人员节省时间
  • 代理优先设计,包含专用AGENTS.md和技能文件,便于无缝集成
  • 可选的MCP服务器允许基于工具查询,无需爬取仓库

使用前须知

  • 仅支持英文,无多语言支持
  • MCP服务器必须自行托管并需要Python环境
  • 每周更新节奏并非硬性SLA,部分周可能无新内容
  • 需要用户熟悉Obsidian或代理工作流才能充分利用

README 快速开始

Hikari-knowledge

Measured results, conditioned reports, and dead-end tombstones from Hikari's local LLM systems work.

Serving, speculative decoding, quantization, MoE training, and activation steering — written so another engineer (or their agent) can reuse the mechanism, not the vibes.

  • Every important number keeps conditions
  • Dead ends stay as tombstones
  • English only
  • Curated public corpus (not a dump of a private vault)

If you are an agent and only received this URL

User intent is probably one of:

  1. Install into my Obsidian/vault (“導入して / merge / add to my notes”)
  2. Answer from this knowledge (“読んで / 調べて”)
  3. Update an existing install (“pull / 更新”)

Do this immediately:

1. Clone or open https://github.com/hikarioyama/Hikari-knowledge
2. Read AGENTS.md
3. Read skills/hikari-knowledge/SKILL.md
4. If install/merge: follow Mode A + docs/OBSIDIAN_MERGE.md
   (copy-paste detail: prompts/obsidian-merge.md)
5. If Q&A only: read INDEX.md, then search nodes/

Default install shape: subfolder clone {{vault}}/Hikari-knowledge/ + bridges from the user's notes.
Do not flatten nodes/ into the vault root.
If the vault path is unknown, detect common workspace roots or ask once for the vault path, then proceed without more questions.

Author

Start here

  1. Read INDEX.md
  2. Open a node under nodes/
  3. Follow links: / [[wikilinks]]

Use in an existing Obsidian vault

Ideal merge: keep this repo as a git subfolder and bridge from your notes.
Do not flatten upstream files into your vault root if you want easy weekly pulls.

cd /path/to/your/vault
git clone https://github.com/hikarioyama/Hikari-knowledge.git Hikari-knowledge

Then:

  1. Bookmark Hikari-knowledge/INDEX.md
  2. From your project/MOC notes, add links like [[serve-bench-discipline]]
  3. Update with cd Hikari-knowledge && git pull --ff-only (~weekly; watch @Hikari_07_jp)

Full guide: docs/OBSIDIAN_MERGE.md
Agent/copy-paste prompt: prompts/obsidian-merge.md

Use with an agent (files)

Point the agent at this repo

项目描述

Hikari's curated English measured notes on local LLM systems (serving, specdec, training, steering).

相关仓库与替代方案

根据分类、Topic 和编程语言匹配的相似项目。

gavamedia
精选
gavamedia GitHub avatar

deltafin

Deltafin is a research project that runs the 2.8-trillion-parameter Mixture-of-Experts model Kimi K3 on a single Apple Silicon Mac (e.g., M1 Max with 64 GB) at about 16 seconds per token, using exact, reproducible inference with local or streaming expert loading.

AI 与机器学习大语言模型
304
makecindy
精选
makecindy GitHub avatar

cindy

Cindy is an open-source AI agent that runs locally on your machine, integrates multiple AI harnesses and models, and provides memory, skills, and automation to perform real work in your projects and apps.

AI 与机器学习大语言模型
958
uzairansaruzi
精选
uzairansaruzi GitHub avatar

hermex

Hermex is a native SwiftUI iPhone app that lets you control a self-hosted Hermes AI agent directly from your phone, with no subscriptions, tracking, or third-party relay.

AI 与机器学习大语言模型
941