Instructions to use SceneWorks/kolors-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use SceneWorks/kolors-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir kolors-mlx SceneWorks/kolors-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
license_name: kolors-model-license
|
| 4 |
+
license_link: LICENSE
|
| 5 |
+
library_name: mlx
|
| 6 |
+
pipeline_tag: text-to-image
|
| 7 |
+
base_model: Kwai-Kolors/Kolors-diffusers
|
| 8 |
+
tags:
|
| 9 |
+
- mlx
|
| 10 |
+
- kolors
|
| 11 |
+
- text-to-image
|
| 12 |
+
- quantized
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# Kolors — MLX quantization tiers (SceneWorks)
|
| 16 |
+
|
| 17 |
+
Pre-quantized MLX turnkey tiers of [Kwai-Kolors/Kolors-diffusers](https://huggingface.co/Kwai-Kolors/Kolors-diffusers)
|
| 18 |
+
for the SceneWorks native (`mlx-gen`) runtime. Each tier subdir is a self-contained snapshot — SDXL-style
|
| 19 |
+
U-Net + **ChatGLM3-6B** text encoder + SDXL VAE + baked fast tokenizer — that loads **packed**, with no
|
| 20 |
+
dense transient and no install-time conversion peak.
|
| 21 |
+
|
| 22 |
+
| Tier | Subdir | Size | Notes |
|
| 23 |
+
|------|--------|------|-------|
|
| 24 |
+
| Q4 (default) | `q4/` | ~6.0 GiB | U-Net + ChatGLM3 packed 4-bit; dense VAE |
|
| 25 |
+
| Q8 | `q8/` | ~9.7 GiB | U-Net + ChatGLM3 packed 8-bit; dense VAE |
|
| 26 |
+
| bf16 | `bf16/`| ~17 GiB | dense 16-bit weights |
|
| 27 |
+
|
| 28 |
+
**Quantization:** group-wise affine, group size 64, weight-only, byte-identical to the runtime's
|
| 29 |
+
load-time seam. The VAE is never quantized (it runs f32). The fast `tokenizer.json` is derived from the
|
| 30 |
+
ChatGLM3 slow tokenizer ([`SceneWorks/kolors-chatglm3-tokenizer`](https://huggingface.co/SceneWorks/kolors-chatglm3-tokenizer))
|
| 31 |
+
and baked into each tier.
|
| 32 |
+
|
| 33 |
+
## License & attribution
|
| 34 |
+
|
| 35 |
+
Model weights are Kolors, © Kwai-Kolors, under the **Kolors Model License Agreement** — see
|
| 36 |
+
[`LICENSE`](LICENSE) (verbatim). Commercial use requires a separate license from the original licensor.
|
| 37 |
+
Re-hosted by SceneWorks for runtime provisioning; the license and its use-restrictions travel with the
|
| 38 |
+
weights and bind downstream recipients (per §3(a) of the license).
|