Still issues with tooling from time to time....
Template v19 works pretty well, but I still do see issues with tooling occasionally 😥
(Using ZED, running Qwen3.6-35b-a3b-Q4km in LM Studio with:
MoE = 12
thinking = off
temperature = 0.7
top_k = 20
repeat = off
presence = off
top_p = 0.95
min_p = 0.05
template = v19)
...and another example:
When thinking is disabled, smaller or quantized models like Q4_K_M frequently struggle to structure tool calls and tend to drop parameters. Try keeping thinking enabled. The new v20 template also adds fallback handling to help keep the context clean when this happens.
Ah, thanks for the info; I did not know this. You truly are awesome 👍
Will try with Thinking too.
Nah; unfortunately I do not think that that is it... 🙁
To add more context -> It seems that mostly it is actually "subagents" who do have issues with tooling, not the "main chat thread".
Do subagents actually inherit the same set of System prompt and Prompt template? 🤔
Hi! The massive v21 update was just released. It completely overhauled tool-calling compatibility (switching to native Hermes JSON for inference engines like llama.cpp and LM Studio), fixed the preserve_thinking amnesia stalls/loops, and resolved several </think> parsing and prompt injection bugs.
This issue should now be fully resolved in v21. I am closing this discussion, but please feel free to reopen or create a new one if you are still experiencing any trouble. Thanks!