Running QwenVL (qwen vision) locally with docker

TLDR

Qwen’s vision models are the best open source vision models available, but are tricky to run locally because they’re not supported by Ollama or llama.cpp. But If you have a linux box with an nvidia gpu you can run run qwen-vl directly by doing:

Experimenting with VS Code’s new Agent mode

In a refreshing change of pace the twitter and youtube crowd has been ooh-ing and ahh-ing over Github Copilot’s new agent mode, with lofty praise like “Github Copilot might be relevant again!” It seems Copilot might finally be catching up to Cursor/Windsurf, and better yet they have a free tier! So I figured I’d try it out once more.

DIY monorepo flow with git subtree

Experimenting with getting a good monorepo setup going. The goal is to have something that’s easy to work with (ie it should feel like I’m only working in one repo), but to have “subrepos” inside it that correspond to different git repos. Basically what git submodules does, but without all the headache.

Achieving the holy grail: vim bindings everywhere

So it turns out it’s possible to get vim bindings everywhere, in software, and sharing the same config on both macs, windows and linux machines. The secret? A cool little tool called keymapper.

Advanced Logseq queries can be useful

Logseq has a nifty feature where if you use their time tracking primitives it’ll automatically use a query under the hood to show you the tasks that you’re currently working on (ie blocks that begin with NOW). I vaguely knew that under the hood it was using a Logseq advanced query, but hadn’t previously looked into how they actually worked, so today I decided to rectify that.

How to use Deepseek R1 and o1 in VS Code

This is definitely in the “experimental” phase and I’m not using it regularly yet, but if you want a Windsurf and/or Cursor like experience but using o1 or Deepseek R1 here’s a quick way to get an 80/20 to get it going