vscode extension

#1
by ludeksvoboda - opened

Discussion how the extension works

Hi,
What is the length of your completions? For me it is sometimes only one token or a word, in few cases it completes whole line. Does it depend on your available compute? This is for CandidateNum set to max (5 slow), bit annoying that there is not more choices. Is there any way to make completions longer?

Knowledge Engineering Group (KEG) & Data Mining at Tsinghua University org

Hi! For the extension, we don't set a fixed length for each generation. Instead, every call will generate until a whole line is completed (until a "\n"). We have also implemented a function-level completion mode, but it will take more time and thus serve fewer users. We are working on accelerating the inference and might add this in the future version.

Oh I see. Thank you for swift reply.
What does CandidateNum sets then? Is it something like a number of generations and then bestfitting one is chosen? I guess I should read the paper :)
Thank you

Knowledge Engineering Group (KEG) & Data Mining at Tsinghua University org

In fact, it is the same thing as the number of generations, but it won't choose the best one for you. Instead, you can use the shortcut 'option' + '[' or ']' (or in windows, 'alt' + '[' or ']') to switch between the candidates. Please see this guidance for more details: https://github.com/THUDM/CodeGeeX/blob/main/vscode-extension/README.md.

Thanks, awesome work btw. Copilot can go and suck it.

Sign up or log in to comment