SuperGemma-4-12b-abliterated - NVF4 / NVFP4 4-bit
This repository contains an MLX NVFP4 4-bit quantization of the BF16 parent model.
Parent model: Jiunsong/SuperGemma-4-12b-abliterated
Parent Benchmark Summary
| Benchmark | Gemma4 12B original | SuperGemma-4-12b-abliterated | Delta |
|---|---|---|---|
| Overall public top-5 500 | 23.8 | 44.6 | +20.8 |
| GPQA Diamond | 10.0 | 19.0 | +9.0 |
| MMLU-Pro | 17.0 | 18.0 | +1.0 |
| IFEval | 61.0 | 59.0 | -2.0 |
| HumanEval+ | 18.0 | 46.0 | +28.0 |
| MBPP+ | 13.0 | 81.0 | +68.0 |
Usage
pip install -U mlx-lm
mlx_lm.generate --model Jiunsong/SuperGemma-4-12b-abliterated-nvf4 --prompt "Write a short release note."
Notes
- This quantized build is provided for smaller local inference deployments.
- The included MLX chat template uses a direct instruction format for clean local text-generation output.
- For benchmark numbers and validation details, see the BF16 parent repository.