Qwick-3.5-9B-MLX-4bit

Qwick-3.5-9B-MLX-4bit

MLX 4-bit quantization of horiuchinobuyuki/Qwick-3.5-9B for Apple Silicon Macs. Text generation only; the vision tower is not included in this MLX build.

Qwick-3.5-9B fine-tunes Qwen3.5-9B for shorter reasoning at comparable quality. See the base model card for the motivation, full benchmark table, and provenance.

Use

pip install mlx-lm
mlx_lm.generate --model horiuchinobuyuki/Qwick-3.5-9B-MLX-4bit \
  --prompt "Explain why the harmonic series diverges." --max-tokens 512

from mlx_lm import load, generate

model, tokenizer = load("horiuchinobuyuki/Qwick-3.5-9B-MLX-4bit")
messages = [{"role": "user", "content": "Explain why the harmonic series diverges."}]
prompt = tokenizer.apply_chat_template(messages, add_generation_prompt=True)
print(generate(model, tokenizer, prompt=prompt, max_tokens=512))

Conversion

Converted from the BF16 checkpoint horiuchinobuyuki/Qwick-3.5-9B with mlx-lm 0.31.3:

mlx_lm.convert --hf-path horiuchinobuyuki/Qwick-3.5-9B \
  -q --q-bits 4 --q-group-size 64

Affine quantization, group size 64, about 4.5 bits per weight. For higher quality at a larger size, see Qwick-3.5-9B-MLX-8bit.

License

Apache-2.0, same as the base model. Derived from Qwen/Qwen3.5-9B (Apache-2.0). See the base model card for attribution and citation.