spoomplesmaxx-27b-4500-mlx-6bit

MLX 6-bit quantized version of aimeri/spoomplesmaxx-27b-4500, converted with mlx-vlm.

About the base model

spoomplesmaxx-27b-4500 is a continued pretraining (CPT) of Google's Gemma 3 27B PT, trained for 4,500 steps on text data using Unsloth. The SigLIP vision encoder is preserved from the original Gemma 3 architecture — the model retains multimodal (text + image) capability. See the base model card for full training details.

Quantization details

Parameter	Value
Bits	6
Group size	64
Mode	Affine

Usage

pip install -U mlx-vlm

Text + image:

python -m mlx_vlm.generate \
    --model aimeri/spoomplesmaxx-27b-4500-mlx-6bit \
    --prompt "Describe this image." \
    --image path/to/image.jpg \
    --max-tokens 200

Text only:

python -m mlx_vlm.generate \
    --model aimeri/spoomplesmaxx-27b-4500-mlx-6bit \
    --prompt "Once upon a time" \
    --max-tokens 200

Python:

from mlx_vlm import load, generate

model, processor = load("aimeri/spoomplesmaxx-27b-4500-mlx-6bit")
output = generate(model, processor, prompt="Describe this image.", image="path/to/image.jpg", max_tokens=200)
print(output)