---
base_model: GestaltLabs/Ornstein-3.6-27B-RYS
language:
- en
license: apache-2.0
pipeline_tag: text-generation
tags:
- gguf
- llama.cpp
- qwen3.5
- qwen3.6
- rys
- text-generation
- quantized
- canada
- sovereign-ai
---

[![Ornstein-3.6-27B-RYS]( Ornstein3.6-27B-RYS.png)

# Ornstein-3.6-27B-RYS-GGUF

GGUF quantizations of [GestaltLabs/Ornstein-3.6-27B-RYS](https://huggingface.co/GestaltLabs/Ornstein-3.6-27B-RYS) — the RYS-enhanced dense Ornstein model.

## About Gestalt Lab

We are a proudly Canadian research collective working to advance **sovereign Canadian AI** — open-weight models that Canadians (and everyone else) can run locally, study, and build on without dependence on closed foreign APIs. All training, fine-tuning, and quantization is done on local and self-funded compute. By supporting this work, you help keep frontier model development accessible, transparent, and under Canadian stewardship.

## Important: requires a patched llama.cpp

RYS duplicates one of the middle layers, which breaks the hardcoded `full_attention_interval = 4` assumption in stock llama.cpp's Qwen3.5 loader. These GGUFs are re-converted with **per-layer `head_count_kv` baked in**, and you need a llama.cpp that reads that per-layer metadata instead of falling back to the interval formula.

**Patched fork:** [https://github.com/DJLougen/llama.cpp](https://github.com/DJLougen/llama.cpp) (default branch `rys-qwen35`, one commit on top of `ggml-org/llama.cpp@d00685831`, fully backward-compatible).

Stock llama.cpp, Ollama, LM Studio, and any other inference runtime built on stock llama.cpp will currently fail to load these files with a `check_tensor_dims` error on `blk.33` — this is expected until/unless the patch is upstreamed.

## Support This Work

Our training compute is entirely self-funded. If this model is useful to you, consider supporting the lab:

**[Support on Ko-fi](https://ko-fi.com/djlougen)**

* * *

## Available Quantizations

| File | Quant | Size | Notes |
|------|-------|------|-------|
| `ornstein-3.6-27b-rys-q8_0.gguf` | Q8_0 | ~27 GB | Near-lossless, largest |
| `ornstein-3.6-27b-rys-q6_k.gguf` | Q6_K | ~21 GB | Very high quality |
| `ornstein-3.6-27b-rys-q5_k_m.gguf` | Q5_K_M | ~18 GB | Strong quality/size balance |
| `ornstein-3.6-27b-rys-q4_k_m.gguf` | Q4_K_M | ~16 GB | Recommended default |
| `ornstein-3.6-27b-rys-q3_k_m.gguf` | Q3_K_M | ~12 GB | Low-memory option |

## Model Lineage

```
Qwen 3.6 27B → Ornstein3.6 (DDM fine-tune) → RYS (layer 33 dup, +49%)
```

## Model Details

* **Architecture:** Qwen3.5 dense
* **Parameters:** ~27B active
* **Layers:** 65 (64 original + 1 RYS-duplicated layer 33)
* **Context:** 131,072 tokens
* **GGUF metadata:** per-layer `head_count_kv` array encoding the RYS-shifted attention pattern

## Usage

### Build the patched llama.cpp

```bash
git clone https://github.com/DJLougen/llama.cpp.git
cd llama.cpp
git checkout rys-qwen35
cmake -B build -DGGML_CUDA=ON -DCMAKE_BUILD_TYPE=Release
cmake --build build -j
```

Drop `-DGGML_CUDA=ON` for a CPU-only build. The patch touches the GGUF loader and three model forward files; backend selection is independent.

### Download + run

```bash
hf download GestaltLabs/Ornstein-3.6-27B-RYS-GGUF \
    ornstein-3.6-27b-rys-q4_k_m.gguf \
    --local-dir .

./build/bin/llama-server \
    -m ornstein-3.6-27b-rys-q4_k_m.gguf \
    --host 0.0.0.0 --port 8080 \
    --n-gpu-layers 99 --ctx-size 131072 \
    --flash-attn on --jinja \
    -ctk q4_0 -ctv q4_0
```

## License

Apache 2.0