Wrote a package for the gpu-poor/mac-poor to run ollama via remote servers (paid, colab, kaggle etc.)

Just 2 lines and your local ollama can run all models on server-side GPU:

> pip install ollama-remote
> ollama-remote

github.com/amitness/oll...

Same energy (h/t @hamel.bsky.social )

x.com/HamelHusain/...

The most interesting part is the filtering and ranking; you can do a bunch of stuff. I went with hackernews ranking for as a start to balance recency vs popularity.

You could even train your own classifiers to make it more personalized; bluesky seems super hackable, love it!

If using OpenAI, you need to make sure the "reasoning" key is before the "answer" fields (i.e. ResponseFormatA if using pydantic)

@dylancastillo.co has done a nice analysis on this and the order matters

dylancastillo.co/posts/llm-py...

Added bluesky to quarto blog. For any @quarto.org users, you can do it via:
1. Save this svg into `images/bluesky.svg`
2. Replace currentColor in 3. Add this yaml code to _quarto.yml