Simon Willison

It looks like the GPT-3.5 error is in `/usr/local/lib/python3.11/site-packages/smol_dev/prompts.py", line 36, in specify_file_paths` That version is installed here: https://github.com/modal-labs/devlooper/blob/4140c87fc29ee16b09300846e371ba6584a8c092/src/main.py#L28-L33 Looks like that package hasn't had an update in a while:...

I'm tempted to switch to running https://huggingface.co/TaylorAI/gte-tiny instead, it's only about 80MB, should run well (and cache the model too) in GitHub Actions. I could use `llm embed-multi` for that.

This plugin is great! Would you be OK renaming the fragment to `docling:` just to avoid any confusion between `doc:` and the `docs:` one from https://github.com/simonw/llm-docs?

This recipe seems to work for Qwen2-VL-2B-Instruct: ```bash python -m mlx_vlm.generate \ --model Qwen/Qwen2-VL-2B-Instruct \ --max-tokens 100 \ --temp 0.0 \ --image django-roadmap.png \ --prompt "Describe image in detail, include...

Supporting this in `uv run` seems like it could be really useful to me. When I'm working with Django I often have it pick up `os.environ["..."]` for things like API...

Alternatively this recipe could use this URL instead: https://github.com/astral-sh/uv/releases/download/0.3.0/uv-installer.sh Which would at least keep all network traffic within GitHub.

`setup-uv` action would be fantastic - or even getting `uv` into the default runner, but that's GitHub's call. I messed around with caching a bit and [got this to work](https://github.com/simonw/playing-with-actions-single/blob/09040ddd30c0a85d03d4d4c472def0a32a23fef6/.github/workflows/uv.yaml):...

An official Astral version of that reusable Action would be really neat. My interest is primarily in avoiding hitting external network resources for repeat runs of my CI builds.

Relevant docs: https://www.sqlite.org/uri.html

8 commits since last release: https://github.com/simonw/sqlite-utils/compare/3.38...8e7d018fa2582e90433c242168994e4b8f504ed8