All errors when using LM Studio

[BSTail]

Known working endpoints error
2026-05-04 17:36:35 [ERROR]
[LM STUDIO SERVER] {
"error": {
"message": "'input' is required",
"type": "invalid_request",
"code": "invalid_union",
"param": "input"
}
}

[awesomez]

After days of struggle, I managed to use qwen3 (official lm studio build)'s jinja text (find it under my models) to work with the "qwen3.6-27b-claude-opus-reasoning-distill-v2@q8_0" model with these NON-qwen default settings:

temperature:1
top_k:25
repeat_penalty:1.05
presence_penalty:0
top_p:1
min_p:0.2

Unified KV (on)
Flash Attn (on)
Try mmap (on)
K Cache Quantization (on) - Q8_0
V Cache Quantization (on) - Q8_0

Context length 120000

NOTE:

Be sure to remove the "temperature:0.2" in the space agent model settings and copy the name of the model in the developer/local server in LM Studio.

These settings so far (after an entire weekend of tearing my hair out) worked for me in Space Agent.

They're still not perfect, but they're far FAR better than anything else I could find.

[awesomez]

Also, very IMPORTANT:

I notice that, when you restart Space Agent, the default "temperature:0.2" in the custom instructions section comes back on occation. This WILL break your model's intelligence if it's set to 0.2 (for example in my model above).

If you notice your model acting incompetently, I bet this is it.