presence_penalty parameter missing from inference parameters panel (required for Qwen 3.6)

[sammcj]

Which version of LM Studio?
Example: LM Studio 0.3.6

• LM Studio 0.4.12+1
• LM Studio MLX (Apple M5) v1.7.0

Which operating system?
macOS

What is the bug?

presence_penalty is an important inference parameter and is rather critical to running Qwen 3.6, however the parameter isn't available in LM Studio's inference settings panel.

As LM Studio is closed source I'm not able to inspect the code and raise a PR to add the parameter myself

• Here is example of it's value and implementation in oMLX: Add presence penalty and min-p inference params jundot/omlx#94 (comment)
• Unsloth also publish recommendations on many models to configure presence_penalty to 1.5 for non-coding tasks https://unsloth.ai/docs/models/qwen3.6#recommended-settings
• PR from when it was supported in mlx-lm: Presence and frequency penalties ml-explore/mlx-lm#971

It looks like this was previously mentioned #1604 but perhaps that was for llama.cpp only?

Screenshots

Logs
N/A

To Reproduce
Steps to reproduce the behaviour:

1. Go to set the presence_penalty to 1.5 for Qwen 3.6
2. Realise there is no presence_penalty parameter in the inference parameters panel

Previously mentioned as implemented in #1604 , but perhaps that was for llama.cpp only?

[Bustycat]

I really don't like the team being so vocal about Qwen and Mac but having done nothing on this for two months.

[AirRunner]

0.4.7 adds llm.prediction.llama.presencePenalty, but nothing for MLX because the backend doesn't support it yet.

lmstudio-ai/mlx-engine#319 adds presence_penalty support to the MLX backend.

Once merged, the Node bridge will need to map llm.prediction.presencePenalty in the kwargs passed to create_generator for it to be available in the UI.