Faen Zhang

CTO @AInnovation
AI First!

[Faen Zhang @faen_zhang]

@Beamsters1 that is will be great!

[Faen Zhang @faen_zhang]

@TokenRouter_US great

[WY @akokoi1]

原始图片和提示词：

根据这张户型图，写一个单文件 HTML 装修设计工具：

1、2D 平面：按图上的尺寸标注（mm）还原墙体、门窗和房间，承重墙和普通墙分开画，四周标注尺寸，并自动算出房间面积。
2、家具：放好一套默认布置，家具按真实尺寸绘制，可以拖动、旋转、改尺寸，靠近墙面自动吸附。
4、交互：测量工具、拆改非承重墙、更换地面材料、撤销/重做、保存到本地、导出图片。
4、3D：用 three.js 生成同一套方案的 3D 场景，支持鸟瞰和第一人称漫游。
5、切换：2D/3D 在同一页面切换，要有流畅的过渡动画，两边数据实时同步。

[Faen Zhang @faen_zhang]

@ddalcu hello

[陈大黄 @realchendahuang]

现在做个个人小工具或者出海小产品，不建议买云服务器、配环境、折腾 Nginx，纯属自己给自己找罪受。

静态前端直接扔给 Cloudflare Pages，连着 GitHub 推送就能自动构建，全球 CDN 一分钱不收。

后端接口全丢给 Cloudflare Workers，毫秒级冷启动，每天几十万次请求的免费额度根本用不完。

数据��直接上 D1，本质上就是开箱即用的边缘 SQLite，不用自己管连接池，更不用天天提防数据库端口被扫。

用户上传的文件和图片丢进 Cloudflare R2，兼容 S3 协议，出网流量费直接免掉。

验证码和通知邮件接个 Resend，每月 3000 封免费额度，写两行代码直接发，不用自己去伺候邮件服务器折腾解析。

整套全托管组合跑下来，每月几十万次请求直接零账单。

省下买服务器的几十块钱是小事，最爽的是不用半夜被叫醒看日志，也不用操心磁盘满不满。

不用伺候机器，小东西安安静静跑在边缘节点，用到地老天荒也不用掏一分钱。

[Tz @Tz_2022]

这个 tm 的居然是 Claude Opus 5.5 徒手用 JavaScript 填色涂鸦动画一气呵成搞出来的？？？😨

[Faen Zhang @faen_zhang]

@ddalcu @Michaelzsguo mlx-serve is great. my m3 max 128GB run qwen-3.8-next-flash very fast. looking forward to running qwen-image-2.1 on mlx-serve

[Faen Zhang @faen_zhang]

@ddalcu I use mlx.serve with qwen 3.8 next flash on my M3 max 128GB, it is really great

[Faen Zhang @faen_zhang]

@Alibaba_Qwen will you open the model weights ?

[Faen Zhang @faen_zhang]

@_LuoFuli so cool to see Mimo is so open, building a great example for industry

[Faen Zhang @faen_zhang]

@_LuoFuli love mimo

[Fuli Luo @_LuoFuli]

Nearly half a year of silence. We spent it studying one problem: how far RL can scale.

MiMo-V2.6 is in the middle of its RL run right now. Three things we scaled: compute (~2B tokens per step, 1568 prompts × 16 rollouts, fully async), environments and harnesses (multi-task agentic RL, mixed across multiple harnesses in one run), and grader compute (agentic in-group credit assignment, with test-case and rubric-based rewards). We'll open-source the details piece by piece over the coming weeks.

Streaming the run: https://t.co/ZSxahzJRju

[Faen Zhang @faen_zhang]

@TheDavidTai cool

[antirez @antirez]

At this point DwarfStar contains many fast fused kernels for important model families: feel free to steal everything you want from there according to the MIT license, in order to improve your own implementation.

[Faen Zhang @faen_zhang]

@StepFun_ai looking forward to openning weighs

[Faen Zhang @faen_zhang]

@ivanfioravanti wow

[Faen Zhang @faen_zhang]

@ashxhart wow, fantastic

[Ash Hart @ashxhart]

I wanted my Mac Studio and two Sparks to work together, so I built a macOS RDMA driver.

Metal and CUDA shared-buffer transfers now pass in both directions.

Here’s why I built MCDMA, what works today, and the 25 experiments I want to try next. https://t.co/gCRYWdioOY

[David Dalcu @ddalcu]

Anyone with 64GB Ram want to test Qwen Flash Next on MLX-Serve ? should work with existing release, but next one will be faster. https://t.co/sAKK5jnoir

[Faen Zhang @faen_zhang]

@PawelHuryn Qwen3.8-Flash is still missing on https://t.co/920QymsZEX