With recent Qwen 3 release, will llama.vim support it soon?

[tudoanh]

I'm currently using Qwen2.5-Coder-7B-Q8_0-GGUF with llama.vim and have been really impressed with its autocomplete capabilities. Seeing all the praise for the new Qwen 3 models, I imagine they could be a significant upgrade. Are there any plans to add support for Qwen 3 in llama.vim in the near future?

[ggerganov]

The new Qwen 3 models are not FIM compatible. Hopefully the Qwen team will release Qwen 3 Coder models.

[usrlocalben]

Qwen3-Coder-30B-A3B-Instruct has been released. It works-for-me as a drop-in-replacement for OP's model.

Tested with unsloth gguf

FIM support is covered by the documentation.

[tudoanh]

Nice, I just tested and it working fine. Very fast also.

[mickmcq]

May I ask how you're using it as a drop-in replacement? Previously, I was successfully running:

llama-server --fim-qwen-7b-default

but when I try

llama-server -hf unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF:Q4_K_M

it starts the main loop but when I run Vim, I just keep getting the message

Job failed with exit code: 7

whenever I enter insert mode and the terminal instance with the llama-server command just remains on the initial update-slots: all slots are idle message.

[tudoanh]

llama-server -hf unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q4_0 -ngl 99 --jinja --temp 0.0 --port 8999

This one working for me

[mickmcq]

Perfect! Thank you!