README.md

---
base_model: microsoft/AesCode-32B
base_model_relation: finetune
library_name: mlx
pipeline_tag: image-text-to-text
tags:
- mlx
- multimodal
- vision
- code-generation
- html
- apple-silicon
- 4-bit
language:
- en
license: apache-2.0
---

# AesCode-32B MLX 4bit

A 4-bit MLX quantization of
[microsoft/AesCode-32B](https://huggingface.co/microsoft/AesCode-32B) for Apple
Silicon, **with the vision tower kept in bf16 so it still sees**.

AesCode-32B is a `Qwen3VLForConditionalGeneration` finetune of
[Qwen/Qwen3-VL-32B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-32B-Instruct)
that takes a screenshot and writes the HTML and CSS to recreate it (slides,
posters, dashboards, product pages). Only the language tower is
quantized to 4-bit; the 351 `vision_tower.*` tensors stay in bfloat16, so the
image path keeps its quality.

**4-bit is about a quarter the size for the smallest machines; bf16 is full quality.**

## Sizes in this family

- [AesCode-32B-MLX-bf16 (full quality)](https://huggingface.co/divinetribe/AesCode-32B-MLX-bf16)
- [AesCode-32B-MLX-8bit (half size)](https://huggingface.co/divinetribe/AesCode-32B-MLX-8bit)
- [AesCode-32B-MLX-4bit (quarter size)](https://huggingface.co/divinetribe/AesCode-32B-MLX-4bit)

## What is in this build

| | |
|---|---|
| precision | 4-bit language tower, bf16 vision tower |
| vision | all 351 `vision_tower.*` tensors kept, `vision_config` and processor configs intact |
| size | about 18 GB on disk |
| framework | runs with `mlx-vlm` on Apple Silicon |

## Vision was tested before upload

Every size of this model is put through the same gate before it ships: a real
screenshot of a live ineedhemp.com product page is fed in and the model is asked
for an SEO, marketing and accessibility audit plus image alt-text. A smaller build
is only published if it still reads the page about as well as the full-precision
one. This 4-bit build passed. From its audit of the real page:

> The page is a product detail for a "Divine Crossing v5 Rebuildable Concentrate Heater" on a dark, vape-themed store. Top navigation bar, a two-column grid: left product imagery with a main photo, zoom icon and four thumbnails; right copy and a purchase form with star rating, price range, trust badges (shipping, warranty, American-owned), a description tab row and social-sharing buttons.

It named the real on-page elements and wrote usable alt-text, so the quantization
did not break its vision.

## Use

```bash
pip install mlx-vlm
python -m mlx_vlm.generate --model divinetribe/AesCode-32B-MLX-4bit \
  --image screenshot.png \
  --prompt "Recreate this screenshot as a single HTML file with inline CSS." \
  --max-tokens 1200
```

Runs on an M-series Mac. No cloud inference, no API keys, no telemetry.

## Benchmarks

The most authoritative evaluation that actually covers this model is the one in
AesCode's own paper, **"AesCode: Aesthetic Code Generation with Decoupled
Cross-Modal Rewards"** ([arXiv:2610.18860](https://arxiv.org/abs/2610.18860)).
There is no independent public leaderboard that lists AesCode specifically, so
these numbers come from the authors' benchmark of 300 infographic-style samples,
scored for rule-following correctness and visual quality. Scores quoted from the
[model card](https://huggingface.co/microsoft/AesCode-32B):

| Model (with reference image) | Overall | Visual |
|---|---|---|
| **AesCode-32B** | **85.89** | 77.44 |
| Claude Opus 4.8 (no reference) | 76.78 | |
| GPT-5.5 (no reference) | 76.31 | |
| Qwen3-VL-32B base (with reference) | 61.10 | |

These are the original model's numbers. For general-vision numbers on the backbone
itself, see the official
[Qwen3-VL](https://huggingface.co/Qwen/Qwen3-VL-32B-Instruct) evaluations (that is
the base model's benchmark, not AesCode's).

## Credits

This build stands entirely on other people's work. Credit where it is due:

- **The AesCode team at Microsoft**, who created the model and the benchmark:
  Xin Gao, Yifan Yang, Xin Li, Pan Wang, Yan Li, Muzhao Tian, Mingxi Cheng, Bei
  Liu, Kai Qiu, Qi Dai, Dong Chen, Yue Dong, Leandros Tassiulas, Rex Ying, Ji Li
  and Chong Luo. Paper: [arXiv:2610.18860](https://arxiv.org/abs/2610.18860),
  model: [microsoft/AesCode-32B](https://huggingface.co/microsoft/AesCode-32B).
- **The Qwen team at Alibaba**, for the
  [Qwen3-VL-32B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-32B-Instruct) base.
- **The MLX team at Apple** and **Prince Canuma** and the
  [mlx-vlm](https://github.com/Blaizzy/mlx-vlm) contributors, for the open-source
  tools that let a model like this run natively and privately on a Mac.

Thank you to all of them.

## Note

Quantization trades a little precision for size. It does not add knowledge and it
does not make output correct. You are responsible for what you build with it.

Built and published by [divinetribe](https://huggingface.co/divinetribe) ·
[more models](https://nicedreamzwholesale.com/software/huggingface/)