DavidAU

/

Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING-MAX-NEOCODE-Imatrix-GGUF

like 85

Tools were tested, and this model has upgraded jinja template; which includes tool updates.
Keep in mind; for tools you should use Q6/Q8 ; as lower quants may cause issues with tool calls/processing.

Update:
Not good enough for tool call even with Q6_K.

==================================================

Tool calling works on Q6_K after Jinja fix, Q4_K_M too weak

Tested both quants via LM Studio API (OpenClaw agent backend). The model's Jinja template might have a bug that breaks all tool/function calls. Here's the fix:

Template fix (LM Studio → My Models → Model Settings → Prompt Template):

-Change
{{- (args_value | tojson | safe if args_value is mapping or args_value is sequence else args_value | string) + '\n\n' }}
+To
{{- args_value | tojson + '\n\n' }}

Results:

• Q6_K: read, write, edit all work reliably after fix
• Q4_K_M: write barely works, read consistently fails with empty arguments

For agent/tool-calling use, pick Q6_K.

Would be great to have the fix baked into the GGUF metadata for future uploads.

Thanks for the model!

Q4_K_M is fine for text-only chat.

Tip: Use {% set enable_thinking = false %} at the top to stop thinking for OpenClaw agents.

I get an error when trying to use any verison with lmstudio/openclaw. Have no issues with other models. Something about jinja templates. Even with the Q8 model.

In LMS ; load the model normally.
Make sure you are in DEV mode; right click the RIGHT LOWER panel -> this will allow you to see/edit the Jinja template.
Copy/PASTE the jinja template from Qwen 3.5 9B REPO to see if the ORG template will solve the tool issue(s).

The same problem appears on lmstudio, and the tool cannot be called.