fudingyu (@fdy) profile picture

Vibe Coding + Vibe Marketing + Vibe Selling, Startups.

Joined March 2009
Photo by fudingyu (@fdy)Photo by fudingyu (@fdy)Photo by fudingyu (@fdy)Photo by fudingyu (@fdy)Photo by fudingyu (@fdy)Photo by fudingyu (@fdy)Photo by fudingyu (@fdy)Photo by fudingyu (@fdy)Photo by fudingyu (@fdy)Photo by fudingyu (@fdy)
X bookmarks are where good content goes to die. No fuzzy search, no topic browsing — so I built xbm, a 100% local bookmark manager: 🔎 FTS5 + EmbeddingGemma hybrid search 🗂 3 category axes: 14,441 → 61 🌌 UMAP map, timeline, threads ledger #LocalLLM #LocalLLaMA #RAG #Gemma
7
3
90
The threads ledger separates algorithm hype from your real knowledge assets: ⚠ interrupted: active 23 months, silent 4 — the feed stopped, you didn't ✓ persistent: active 44 months, spanning 13 years 🌱 new shoots · ⚡ hot pulses #SecondBrain #SelfHosted
9
Timeline view: monthly topic stats across all bookmarks, as a narrative of how your focus moved. 2012 knowledge → 2020 life abroad → 2021-22 AI tooling → 2023+ "AI monetization" as the main thread. Every node says what you turned to, and why. #BuildInPublic #PKM
38
UMAP projection of all 14,386 bookmarks — your interests finally have a shape: 🌌 clusters = what you really care about 🎯 isolated points = long-tail the algo never showed you 🔁 dense overlaps = topics you keep re-saving Invisible in any list view. #DataViz #PKM
27
Problem 2: no structure, just a timeline. xbm slices bookmarks along 3 orthogonal axes: 📁 12 zero-shot categories 🔤 1,115 keyword clusters 🧠 27 LLM semantic clusters Intersection = semantically close AND same wording: 14,441 → 61. #LocalFirst #DevTools
24
X only matches literal words: "inference optimization" never finds the CUDA tweet that never says those words. xbm runs two retrievers in parallel, fused with RRF: 🔤 FTS5 trigram, CJK substring match 🧠 local EmbeddingGemma cosine #SemanticSearch #RAG
9
程序员节ZCode送了3亿Token,于是做了个工具管理X书签,以往搜索错一个字都找不到,更别说按语义相似搜索,只能一直往下翻浪费时间。 功能特点: ✅ 语义搜索:搜「推理优化」, ✅ 分类检索:通过关键词和语义条精准收窄 ✅ UMAP 星云图 + 时间脉络 + 线索账本 ✅ 本地 EmbeddingGemma,毫秒级搜索响应
4
3
139
加了个时间脉络的整理: 按月聚合全部书签的主题统计,还原你的认知轨迹: 2012 知识科普 → 2020 海外生活 → 2021-22 AI 工具萌芽 → 2023 后「AI 变现实操」成为主线。 每个节点都写明了转向了什么、为什么转向。
1
15
按月统计书签,得到的只是算法的镜子。这个视图把「算法热点」和「你的知识资产」彻底分开: ⚠ 中断线索:曾活跃 23 个月、沉寂 4 个月——算法早就不推了,但你认真收藏过; ✓ 持续线索:活跃 44 个月、跨 13 年——你的知识资产骨架; 🌱 新芽 · ⚡ 热点脉冲。 它回答的问题是:哪些值得重新捡起来。
11
Unlocked the pure NPU on my Ryzen 7 8845. 🤖⚡ Now running on XDNA INT8: • YOLOv8m ResNet Zero dGPU. Zero cloud. Constraint: PHX/HPT XDNA = CNN INT8 only. LLMs need BF16. Next step: rewrite drivers + operators for BF16. #RyzenAI #AMD #NPU #EdgeAI #LocalLLM #LocalLlama
1
2
79
Currently running on the NPU: YOLOv8m — Object Detection ResNet-50 MobileNetV2 — Classification RetinaFace — Face Detection RCAN 2× — Super-Resolution PHX/HPT XDNA only exposes CNN INT8 today. LLMs want BF16, So need rewriting the driver and operator layer to LLMs running on XDNA
34
I ran the Laya model on llama.cpp to test the game of Snake, and the results were excellent. The speed was fast, and even on my laptop with an AMD Radeon 780M Graphics, it ran smoothly. Thanks @chenchengpro game code. #Jev #Laya #TypeSafe #llamacpp #LocalLLM
1
1
4
364
We’re rolling out some big improvements to Gemma 4, fueled by incredible community feedback and contributions! Here is a breakdown of what’s being fixed and updated in this release: 🧵👇
129
319
88
3,294
721,696
WSL container is now available for public preview. That means you can create, run, test, and debug Linux containers directly through WSL on Windows. Here’s how to get started + a few ways to try it 🧵
33
290
14
1,167
98,781
Introducing InfiniteDiffusion, my independent paper accepted to #SIGGRAPH2026! I have one RTX 3090 Ti. No funding, advisors, or team. By day I'm a new grad SWE at Walmart. The paper has two main contributions: - InfiniteDiffusion: a new approach to infinite generation with diffusion models. - Terrain Diffusion: the world’s first learned procedural terrain generator. Here’s why this matters, and how they are connected. 🧵
152
621
91
6,272
946,677