Single translations always translates to Chinese?

[chasegiunta]

Seems like every time I use the AI button on a single translation, it never takes the locale into account and defaults to an Asian derivative language. When I use the batch translation feature, it seems to work correctly. Any ideas?

[akinoccc]

I can't reproduce it😥

[akinoccc]

500097016-e46ceb5f-2341-4464-b359-03ac3ee0b3e2.mov

[chasegiunta]

Not sure if it matters, but the locales I'm working with are EN & ES. Does it solely use the locale code for figuring out what to translate to?

[akinoccc]

Not sure if it matters, but the locales I'm working with are EN & ES. Does it solely use the locale code for figuring out what to translate to?

Yep, the locale code will be sent to the ai and tell it which is target language

[chasegiunta]

So weird

[chasegiunta]

If I try a different model:

[akinoccc]

If I try a different model:

I recommend using a model larger than 70B because this plugin sends the code context together with the request to the AI. If the model is too small, it may become less effective when handling larger contexts.

[akinoccc]

You can try using the full models to see whether the result improves. I previously tested smaller models as well, but the results were too poor. That’s why I’m now using the DeepSeek 671B, and this issue does not occur with it.

[chasegiunta]

DeepSeek 671B is a huge model! 404GB!

Maybe there needs to be a way to pass less context or specify custom context. It's difficult to fathom that simple string translation would require such huge resources.

[akinoccc]

DeepSeek 671B is a huge model! 404GB!

Maybe there needs to be a way to pass less context or specify custom context. It's difficult to fathom that simple string translation would require such huge resources.

Ha ha… I’m using the cloud API key, but I think at least the 70B model is required. I’m also considering adding an option to decide whether the code should be taken as context. Note that if the code context is removed, the translation may become significantly less accurate.

[chasegiunta]

Another option would be to hook into the GitHub Copilot APIs, as it already has code context.

[akinoccc]

Another option would be to hook into the GitHub Copilot APIs, as it already has code context.

To be researched...Thx

[akinoccc]

Another option would be to hook into the GitHub Copilot APIs, as it already has code context.

Sry, I‘m not found the way to reuse its capability

[akinoccc]

DeepSeek 671B is a huge model! 404GB!

Maybe there needs to be a way to pass less context or specify custom context. It's difficult to fathom that simple string translation would require such huge resources.

And the ext only sends the context of the code file where the target text to be translated is located; it does not send the entire project’s code to the AI.