Model name for custom OpenAI in GUI-Subtrans

[dgoryeo]

Is there a way to set the model name in GUI-Subedit for a custom OpenAI?
I am trying to use DeepSeek v3.

Thanks

[Neoony]

You have to use the local model option.
Some models are filtered out in the OpenAI option

had a similar question: #198

[dgoryeo]

Thank you!

[machinewrapped]

The Deepseek API is fully OpenAI-compatible, so it is just the fact that the model names are filtered to only match gpt models that prevents the Deepseek URL being used with the OpenAI provider.

I'll push out a quick update this weekend to return the model list unfiltered if there are no gpt models at the provided endpoint!

As @Neoony says, the local model option does work with any OpenAI-compatible endpoint too though :-)

In my very quick test Deepseek's translation seems a bit worse than the other main models - more literal, less "flow" and a few muddled lines. However, my test subtitles cost about $0.003 to translate so pretty good value!

[dgoryeo]

Thank you. Yes, Deepseek pricing is quite a good value at the moment.

[machinewrapped]

New release, v1.0.2:
https://github.com/machinewrapped/gpt-subtrans/releases/tag/v1.0.2

I updated the OpenAI provider so that if there are no GPT models in the model list it will return the full list (rather than an empty list), so it can be used with other API compatible providers like DeepSeek.

However, I've also added DeepSeek as an official provider so you can set a separate API key, so probably easier to use that ;-)

Let me know if you encounter any problems!

[dgoryeo]

Thank you so much. I'll test it duirng the weekend.

[dgoryeo]

Hi, I have been some tests with Mistral large model and it works quite well.
I did couple of quick tests with Deepseek but the translation stops midway. I'm wondering if it is caused by their strict moderation policy.

[machinewrapped]

Hmm, could be - I tried to translate a Taiwanese wuxia and mentioned that it was Taiwanese in the description - no result. When I removed that part of the description it did a great job of translating the actual subtitles though!

It could just be that their servers are overloaded though, given how much interest there has been in them this week.