Bugged out qwen models

[Cdingram]

Has anyone else noticed template issues with some qwen 3 variants. These two for example qwen3 1.7B and qwen3-2507 both error for me on the second message if jinja is set to true in the completion options. The non thinking version of qwen3-2507 works fine.

The error is something like "Value is not callable: null at row 48, column 85" and then shows some line of code about stripping think tags.

I'm pretty sure it isn't actually this library but I'm trying to rule out something here vs llama.cpp vs the GGUF provider.

[bernardobelchior]

Yeah, I'm having issues with Qwen/Qwen2.5-3B-Instruct-GGUF/qwen2.5-3b-instruct-q3_k_m.gguf.

Here's a small investigation I've done: https://discord.com/channels/426714625279524876/1368813641477914685/1470521644379279402