regular completion overrides LLM based completion

[tfriedel]

I was trying the demo at https://jtp.io/jupyterlab-codeium/lab/index.html Surprisingly enough, there's an API key already filled in, so I didn't fill in any. After typing a bit, some ghost text started to appear. However I couldn't autocomplete it by pressing tab, 90% of the time.

The problem was, there's often a dropdown with some keywords shown and pressing TAB completes the top keyword from that list, rather than the ghost text. If by some chance there is no dropdown menu shown, pressing tab completes the ghost text.

[jtpio]

Thanks @tfriedel for the report.

This may be fixed by https://github.com/jupyterlab/jupyterlab/pull/16613. Would you be able to try the extension with the latest JupyterLab 4.3.0 pre-release? With something like pip install --pre jupyterlab jupyterlab-codeium.

[tfriedel]

Thanks! I tried this, but stumbled into a bunch of problems:

• no easy way to get an api key. I ended up copy pasting the one from the link above. I was searching in my account on codeium.com for some api key in settings, like I'm used to for other services.
• autocomplete with tab didn't work. If I saw a completion, it showed I can press alt-end. This works but is super awkward to press on my kinesis keyboard. Would prefer TAB and a way to configure it.
• sometimes completions didn't show. E.g. I start typing "os.p" and would expect os.path... Maybe they didn't show because the regular completion still overrides the LLM one? Because for os.p there is a regular completion.