Is it impossible to use qwen3, or other models that don't have <|fim_prefix|> etc. in official template?

[akavel]

I'm only starting to learn all this LLM stuff, so a lot of this is not obvious and confusing to me, please forgive me!

I just discovered a qwen3-coder:30b model on ollama, and it seems to be working quite well for me with CodeCompanion.nvim for chat - much better than other models I tried before now (including some qwen2.5 ones). Unfortunately, when I tried to use it with minuet-ai.nvim, I got a message akin to: "qwen3 does not support insert". From some attempts at searching and reading the minuet-ai docs, I seem to understand it may just not be possible to use it in this case - for example, the "template" at https://ollama.com/library/qwen3-coder:latest/blobs/c6a614465b37 does not seem to have any "<|fim_prefix|>" markers at all. Do I understand it correctly? Is it completely impossible for me to use qwen3 with minuet-ai for code completion?

Or, is there still some way?

[milanglacier]

Check this:

#125 (comment)

Sometimes ollama does not provide the template for <|fim_prefix|> even though the model actually support it (for example qwen3-coder). In that case, you can try the example I provided.

However, for model that does not support FIM completion at all (which is the case for most of the chat model, say, GPT-OSS), it is not possible to use the Openai-FIM-Compatible provider.