---
license: apache-2.0
base_model: google/gemma-4-E4B-it
tags:
  - gguf
  - gemma
  - llama.cpp
  - lmstudio
  - korean
  - fine-tuned
---

# Gemma MLX Studio

GGUF exports for the `best_4b_v2_stronger` run produced with Gemma MLX Studio.

## Files

- `best_4b_v2_stronger-Q4_K_M.gguf`
  - recommended for practical LM Studio / llama.cpp use
- `best_4b_v2_stronger-f16.gguf`
  - high precision archive / conversion source

## Notes

- Base family: `Gemma 4 E4B IT`
- Fine-tuning workflow: local MLX + LoRA, then fused export
- Primary language: Korean

## Suggested LM Studio settings

- Temperature: `0.2`
- Top P: `0.9`
- Top K: `40`
- Repetition Penalty: `1.1`

## Prompt style

This model tends to work best with:

- short and structured instructions
- explicit output length constraints
- proper noun preservation

Example:

```text
First Fire Horizon와 Nightglass Relay의 차이를 3문장으로 설명해줘.
First Fire Horizon에는 튜토리얼, 기본 보급, 첫 항로 개방만 연결하고,
Nightglass Relay에는 신호, 기록, 중계만 연결해.
고유명사는 번역하지 마.
```