RuiSumida
/

Qwen3.5-4B-6bit-text

Model card Files Files and versions

Qwen3.5-4B-6bit, text-only

The weights of mlx-community/Qwen3.5-4B-6bit with the vision tower removed.

All 924 language_model.* tensors are byte-identical to the source (verified with mx.array_equal after re-saving).
The 297 vision_tower.* tensors (667 MB) are dropped. mlx-swift-lm's text Qwen35 model discards those keys on load anyway, so text generation is unchanged; this repack only saves the download and disk space (4.09 GB → 3.42 GB).
config.json, tokenizer files and chat_template.jinja are copied unchanged. Image input is not supported.

Built for the Luke iPhone app (on-device chat, MLX).

Downloads last month: 87

Safetensors

Model size

4B params

Tensor type

U32

·

BF16

·

F32

·

MLX

Hardware compatibility

Log In to add your hardware

6-bit

Inference Providers NEW

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

Model tree for RuiSumida/Qwen3.5-4B-6bit-text

Base model

mlx-community/Qwen3.5-4B-6bit

Quantized

(1)

this model