# LM Studio Setup for `lime-qwen35-9b-lime-reasoning-power-f16-q6_k.gguf`

Official references:

- Prompt Template: https://lmstudio.ai/docs/app/advanced/prompt-template
- model.yaml: https://lmstudio.ai/docs/app/modelyaml

## What This Folder Contains

- `model.yaml`
- `chat_template.jinja`
- `lime-qwen35-9b-lime-reasoning-power-f16-q6_k.gguf`
- `README.md`
- `training_metadata.json`
- `tokenizer_config.json`

## Recommended Usage

### Option 1: Import the GGUF directly into LM Studio

Import:

```text
lime-qwen35-9b-lime-reasoning-power-f16-q6_k.gguf
```

Then manually set the prompt template if LM Studio does not detect it correctly.

### Option 2: Publish a Hugging Face GGUF repo and point `model.yaml` at it

The current `model.yaml` is written for a likely repo name:

```text
naksyu/lime-qwen35-9b-lime-reasoning-power-gguf
```

If you publish under a different repo name, update `model.yaml` before distributing the folder.

## Prompt Template

This model uses:

```text
/think ... </think>
```

It may also emit:

```text
<|lime_think_more|> ... <|lime_end_think|>
<|lime_final|>
```

If reasoning looks broken, empty, or malformed, paste `chat_template.jinja` manually in:

```text
My Models > gear > Prompt Template > Jinja Template
```

## Thinking Toggle

The included `model.yaml` wires:

```text
Enable Thinking -> enable_thinking
```

When disabled, the template opens and immediately closes the initial `/think` block before the visible answer.

## Suggested Default Load

- Context length: `12288`
- Temperature: `0.8`
- Top P: `0.95`
- Top K: `20`

## Important Note

This folder includes the `q6_k` build. Other variants such as `q8_0` and `q4_k_m` exist under the repo-level `gguf/` directory but are not bundled here by default.