Qwen3.5-35B-A3B-8bit (MLX)

This repository provides an MLX quantized conversion of Qwen/Qwen3.5-35B-A3B, prepared for local inference on Apple Silicon.

This is an unofficial conversion published by henrybravo. All model capabilities, limitations, and licensing inherit from the original Qwen release.

Model summary

• Base model: Qwen/Qwen3.5-35B-A3B
• Architecture: Qwen3_5MoeForConditionalGeneration (qwen3_5_moe)
• Quantization: affine, group size 64, weights at 8-bit
• Weight format: safetensors (8 shards)
• Intended runtime: mlx-vlm / MLX ecosystem on macOS

Why this variant

This 8-bit build prioritizes output quality and fidelity over speed.

• Typical local memory target used in router config: about 40 GB
• Local default tested in this setup: max_tokens = 16384
• Expect slower generation than the 4-bit variant

Install

pip install -U mlx==0.31.0 mlx-lm==0.31.0 mlx-vlm==0.4.0 transformers==5.3.0

For OpenAI-compatible local serving and routing on top of MLX models, see mlx-router.

Quick usage (mlx-vlm)

You can run the examples below directly with mlx_vlm.generate, or serve the model through mlx-router.

Text prompt

python -m mlx_vlm.generate \
  --model /path/to/Qwen3.5-35B-A3B-8bit \
  --max-tokens 100 \
  --temperature 0.0 \
  --prompt "Hello, what model are you?"

Vision prompt

python -m mlx_vlm.generate \
  --model /path/to/Qwen3.5-35B-A3B-8bit \
  --max-tokens 200 \
  --temperature 0.0 \
  --prompt "Describe this image in detail." \
  --image https://upload.wikimedia.org/wikipedia/commons/thumb/a/a7/Camponotus_flavomarginatus_ant.jpg/320px-Camponotus_flavomarginatus_ant.jpg

Conversion command used

mlx_vlm.convert \
  --hf-path Qwen/Qwen3.5-35B-A3B \
  --mlx-path ~/models/models--henrybravo--Qwen3.5-35B-A3B-8bit \
  -q \
  --q-bits 8 \
  --dtype bfloat16

Notes

• This repository stores MLX-converted weights; it is not a standard Transformers checkpoint.
• For best official guidance, benchmarks, and serving docs, see the upstream Qwen model card.

Upstream references

• Base model: https://huggingface.co/Qwen/Qwen3.5-35B-A3B
• Original license file: https://huggingface.co/Qwen/Qwen3.5-35B-A3B/blob/main/LICENSE
• Qwen blog: https://qwen.ai/blog?id=qwen3.5