spoomplesmaxx-thrasher-24B — MLX 4-bit

4-bit MLX quant of spoomplesmaxx-thrasher-24B ("Thrash Metal", mimids 02) for Apple Silicon. ~13 GB — comfortable on a 24GB Mac, workable on 16GB with a short context. ChatML template embedded in both tokenizer_config.json and chat_template.jinja.

pip install mlx-lm
mlx_lm.chat --model aimeri/spoomplesmaxx-thrasher-24B-mlx-4bit

Sampler (swept on the full-precision model): temperature 1.0 · min_p 0.05. A mild repetition_penalty 1.05 eliminated the verbatim-loop tail in our sweep at the cost of a rare unfinished turn — a reasonable opt-in. If you want the higher fidelity, the 6-bit (~18 GB) is the quality pick. See the main card for the full story.

For adults. Stays in character by design; bring your own moderation. Apache 2.0.

Downloads last month
20
Safetensors
Model size
4B params
Tensor type
BF16
·
U32
·
MLX
Hardware compatibility
Log In to add your hardware

4-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for aimeri/spoomplesmaxx-thrasher-24B-mlx-4bit

Collection including aimeri/spoomplesmaxx-thrasher-24B-mlx-4bit