search

C-Loftus / talon-ai-tools

Query LLMs and AI tools with voice commands
MIT License
39 stars 14 forks source link

Strip markdown that is output despite our prompting #71

Closed C-Loftus closed 3 weeks ago

C-Loftus commented 3 weeks ago

Sometimes with the model responses, even though we explicitly tell it not to output markdown syntax, it still does it. This PR uses a regex to remove the wrapper.

It also implements pytest since this is just a pure function and thus can be tested more easily.

C-Loftus commented 3 weeks ago

@jaresty I think this should implement the markdown strip we were looking for. Would you mind testing this out a bit and see if you run into the markdown issue? I am no longer getting it but wasn't sure your context or if specific commands were reliably causing markdown syntax and overriding the prompt.

Ignore the branch name, there was a rebase issue but it is fine now and won't be an issue when this is merged.

jaresty commented 3 weeks ago

I'm happy to work off this one-do you think we could integrate the blend changes here? I'm using that one a bit so that's holding me back from switching to this branch.

C-Loftus commented 3 weeks ago

I'm happy to work off this one-do you think we could integrate the blend changes here? I'm using that one a bit so that's holding me back from switching to this branch.

Hmmm we can although it might just be easier to try doing gh pr checkout 71 and then testing a few things quickly. I think the other PR is blocked a bit. Open to ideas if you think otherwise though.

jaresty commented 3 weeks ago

I gave it a shot on a few of my prompts and I did not see the markdown code blocks so it seems to be working based on my cursory checking right now.

C-Loftus commented 3 weeks ago

Ok, will merge this and then we can get the other pr branch up to date. Shouldn't be any regressions but if so I can just revert.