license: mit
base_model: huihui-ai/Huihui-Ornith-1.5-35B-A3B-abliterated
base_model_relation: quantized
library_name: mlx
pipeline_tag: text-generation
tags:
- mlx
- ornith
- abliterated
- text-only
language: - en
Ornith 1.5 35B-A3B abliterated, MLX 3-bit
Text-only MLX 3-bit weights for a 24 GB Apple Silicon Mac. The three safetensors shards are about 14 GB.
Use them with jjakemaness/ornith-omlx-24gb. That repo installs oMLX 0.7.3 and Pi 1.5.0 and keeps these pins:
| Setting | Value |
|---|---|
| Process memory ceiling | 18 GB |
| Concurrent requests | 1 |
| Startup cache blocks | 32 |
| Context window | 65,536 tokens |
| Max output | 32,768 tokens |
| Server | http://127.0.0.1:8000/v1 |
| Model id | Ornith-1.5-35B-A3B-Abliterated-Q3 |
The checkpoint's native context is 262,144 tokens. That window does not fit this memory ceiling, so the oMLX model settings cap it at 65,536.
Lineage: ornith-ai/Ornith-1.5-35B-A3B, then the abliterated release huihui-ai/Huihui-Ornith-1.5-35B-A3B-abliterated. This copy is text-only (qwen3_5_moe). The license follows the upstream MIT model card.
Abliteration reduced refusal behavior. You are responsible for how you use the model.