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
language:
- en
license: apache-2.0
---

# AesCode-32B MLX bf16

A faithful, full-precision (bfloat16) MLX conversion of
[microsoft/AesCode-32B](https://huggingface.co/microsoft/AesCode-32B) for Apple
Silicon, **with the vision tower kept intact**.

AesCode-32B is a `Qwen3VLForConditionalGeneration` finetune of
[Qwen/Qwen3-VL-32B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-32B-Instruct)
that turns a prompt plus a reference image into information-rich visual artifacts
(slides, posters, dashboards, product pages) as HTML and CSS. It sees a screenshot
and writes the code for it.

This conversion changes nothing about the model's behavior. It is the same
33.4B-parameter bf16 weights, re-sharded into MLX format so they load and run
natively on an M-series Mac with no PyTorch and no cloud.

**8-bit is about half the size for smaller 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)

Each size was put through the same ineedhemp.com product-page vision gate before it
shipped. For the uncensored variants, see the
[abliterated line](https://huggingface.co/divinetribe/AesCode-32B-abliterated-MLX-bf16).

## What is in this build

| | |
|---|---|
| precision | bfloat16 (no quantization) |
| vision | all 351 `vision_tower.*` tensors kept, `vision_config` and processor configs intact |
| framework | runs with `mlx-vlm` on Apple Silicon |
| size | about 62 GB on disk |

## Vision was tested before upload

A lot of MLX vision-language uploads are made with tools that only understand the
language half, so the vision tower gets silently dropped and the model fails on
the first image. This one was checked the hard way: a real screenshot of a live
ineedhemp.com product page was fed in and the model was asked for an SEO,
marketing and accessibility audit plus image alt-text.

It read the real page. From its audit:

> SEO: The page needs a clear title tag, probably "Divine Crossing v5 Rebuildable
> Concentrate Heater - Divine Tribe" plus meta description. The breadcrumb and H1
> are good, but adding schema markup for product, review, and offer would help
> rich results.
>
> Marketing/Conversion: Strong points are the bold yellow CTA, clear price range,
> star rating, and benefit-driven copy. The "just the cup, $8.42" section is
> useful but visually buried.
>
> Alt-text: Main product image: "Divine Crossing v5 rebuildable concentrate heater
> with ceramic cup and metal body." Thumbnail 2: "V5 logo and geometric design."

It named the product, the breadcrumb, the star rating, the price range, the yellow
add-to-cart button and a small pricing note buried far down the page, and wrote
usable alt-text for each product photo. The vision path works.

## Use

```bash
pip install mlx-vlm
python -m mlx_vlm.generate --model divinetribe/AesCode-32B-MLX-bf16 \
  --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 straight 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 | |

AesCode-32B lifts visual quality over the same-scale Qwen3-VL baseline by 22.4
points and lands in the range of the closed frontier models. For broader
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
  that AesCode is built on.
- **The MLX team at Apple** and **Prince Canuma** and the
  [mlx-vlm](https://github.com/Blaizzy/mlx-vlm) contributors, whose open-source
  work makes it possible to run a model like this natively and privately on a Mac.

Thank you to all of them.

## Note

This is a faithful conversion for Apple Silicon. It does not add knowledge and it
does not make output correct. You are responsible for what you build with it.

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