Erik Bjäreholt

Improved the eval harness quite a bit in #90, among other changes (incl a lot of Docker stuff). I'm now 80% happy with the harness and am trying to think...

@gptme Optimize the prompt in gptme/tools/patch.py to make it comply better. We want to make sure patches are concise and reliable. We need to be careful not to over-optimize the...

I partially improved this by stopping generation on the ending `>>>>>> UPDATED` in a patch block. Works a lot better now (before it kept going, filling in the rest of...

I've found that Claude 3.5 Sonnet doesn't really suffer from this anymore. Feels like a thing I cannot control anyway, it's all in the hands of the LLM. Trying to...

@gptme patch .github/workflows/bot.yml to implement this. Read the log from the `.jsonl` file. Each line has role and content keys. When the log is written to the issue inside the...

@gptme now check if there's something left to do, apply patches if so

I didn't go all the way to single-letter shortcuts, but we now have `/py` and `/sh`, or any other valid file extension for a supported executable codeblock (not many others,...

It is a lower priority :) It's not that much of a distraction, the changes required would improve the code overall anyway. The web UI is nice for browsing past...

I think these are good ideas, and could easily be added (and would get merged). But right now I'm personally focusing on better general ways to handle context size, see...

@casaout I will follow this with interest (due to my involvement in @ActivityWatch). Do you know when you will start building it?