Downloads · 30 days
105
27% of all-time downloads
L3-8B-Sunfall-v0.5-Stheno-v3.2-4bit-mlx
leonsarmiento/L3-8B-Sunfall-v0.5-Stheno-v3.2-4bit-mlx
L3-8B-Sunfall-v0.5-Stheno-v3.2-4bit-mlx is a machine learning model from leonsarmiento. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for mlx. The card lists the license as unknown.
Uniform 4-bit MLX quantization of Vdr1/L3-8B-Sunfall-v0.5-Stheno-v3.2 — a Llama-3-8B roleplay/creative-writing merge combining the Sunfall v0.5 and Stheno v3.2 lineages.
All-time downloads
386
Public
Parameters
8B
4.5 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors4.5 GB · 100%
Parameter types
How the weights are stored.
U328B · 100%
Base models
Model card
From the Hugging Face model README
L3-8B-Sunfall-v0.5-Stheno-v3.2 — 4-bit (MLX)
Uniform 4-bit MLX quantization of Vdr1/L3-8B-Sunfall-v0.5-Stheno-v3.2 — a Llama-3-8B roleplay/creative-writing merge combining the Sunfall v0.5 and Stheno v3.2 lineages.
Content note: the source model is gated Not-For-All-Audiences (NSFW-capable RP model). Follow the source model's usage terms.
Quantization
| Recipe | Uniform 4-bit, group size 64, affine |
| Bits per weight | 4.5 |
| Size | 4.2 GB |
| Shards | 1 |
| Framework | mlx_lm (native llama) |
Vision/audio: none (text-only model). Chat template shipped in both tokenizer_config.json and chat_template.jinja.
Usage
Works in LM Studio, oMLX, and mlx_lm:
import mlx_lm
model, tokenizer = mlx_lm.load("leonsarmiento/L3-8B-Sunfall-v0.5-Stheno-v3.2-4bit-mlx")
Suggested starting settings (Stheno-lineage RP models; tune to taste — the source card is authoritative): temp 0.8–1.0 · min_p 0.05–0.1 · top_p 1.0 · rep penalty 1.0–1.05 · 8K context
An 8-bit uniform variant is available at leonsarmiento/L3-8B-Sunfall-v0.5-Stheno-v3.2-8bit-mlx.
- Library
- mlx
- Type
- llama
- License
- unknown
- Created
- Sep 4, 2026
- Updated
- Sep 4, 2026