Instructions to use GestaltLabs/Ornstein-3.6-27B-RYS-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use GestaltLabs/Ornstein-3.6-27B-RYS-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="GestaltLabs/Ornstein-3.6-27B-RYS-GGUF",
	filename="ornstein-3.6-27b-rys-q2_k.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use GestaltLabs/Ornstein-3.6-27B-RYS-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M

Use Docker

docker model run hf.co/GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use GestaltLabs/Ornstein-3.6-27B-RYS-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "GestaltLabs/Ornstein-3.6-27B-RYS-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "GestaltLabs/Ornstein-3.6-27B-RYS-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M

Ollama
How to use GestaltLabs/Ornstein-3.6-27B-RYS-GGUF with Ollama:
```
ollama run hf.co/GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M
```

Unsloth Studio new

How to use GestaltLabs/Ornstein-3.6-27B-RYS-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for GestaltLabs/Ornstein-3.6-27B-RYS-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for GestaltLabs/Ornstein-3.6-27B-RYS-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for GestaltLabs/Ornstein-3.6-27B-RYS-GGUF to start chatting

Pi new

How to use GestaltLabs/Ornstein-3.6-27B-RYS-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use GestaltLabs/Ornstein-3.6-27B-RYS-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M

Run Hermes

hermes

Docker Model Runner
How to use GestaltLabs/Ornstein-3.6-27B-RYS-GGUF with Docker Model Runner:
```
docker model run hf.co/GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M
```

Lemonade

How to use GestaltLabs/Ornstein-3.6-27B-RYS-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull GestaltLabs/Ornstein-3.6-27B-RYS-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Ornstein-3.6-27B-RYS-GGUF-Q4_K_M

List all available models

lemonade list

DJLougen commited on 25 days ago

Commit

ef339e8

verified ·

1 Parent(s): c352f80

Update model card: add Canadian lab mission, Ko-fi, patched llama.cpp fork, quant table

Browse files

Files changed (1) hide show

README.md +42 -100

README.md CHANGED Viewed

@@ -1,132 +1,74 @@
----
-base_model: GestaltLabs/Ornstein-3.6-27B-RYS
-base_model_relation: quantized
-tags:
-- gguf
-- llama.cpp
-- qwen3_5
-- qwen3.6
-- multimodal
-- rys
-- quantized
-- 8-bit
-- 6-bit
-- 4-bit
-- 3-bit
-- 2-bit
-license: apache-2.0
-language:
-- en
----
-![Ornstein-3.6-27B-RYS](ChatGPT%20Image%20Apr%2024,%202026,%2006_52_24%20PM(1).png)
 # Ornstein-3.6-27B-RYS-GGUF
-GGUF quantizations for **Ornstein-3.6-27B-RYS** — the RYS-modified variant of Ornstein-3.6-27B with 66 layers (layers 22 & 23 duplicated, zero weight change).
-See the base model repo for full details: **[GestaltLabs/Ornstein-3.6-27B-RYS](https://huggingface.co/GestaltLabs/Ornstein-3.6-27B-RYS)**
----
-## What is RYS?
-**Repeat-Your-Self (RYS)** — Ng, David Noel (2026) — is a zero-training architecture modification for deep transformers. By duplicating a contiguous slice of layers, the model revisits an earlier representation mid-pass, effectively deepening the network without changing any weights.
-For Ornstein-3.6-27B, the optimal config is **i=22, j=24**:
-```
-new_layer_order = [0..23, 22, 23, 24..63]   →   66 layers total
-```
-**Sweep results**: 2,080 configs evaluated, combined delta = **+0.3223** (math +0.010, IFO +0.312).
----
-## Available Quants
-| Quant | Size | BPW | VRAM Required | Best For |
-|-------|------|-----|---------------|----------|
-| **Q8_0** | ~27.4 GB | 8.0 | ~32 GB | Maximum quality, near-lossless |
-| **Q6_K** | ~21.1 GB | 6.0 | ~26 GB | Strong quality, large context |
-| **Q4_K_M** | ~15.8 GB | 4.0 | ~20 GB | **Balanced default**, 24 GB cards |
-| **Q3_K_M** | ~13.0 GB | 3.0 | ~17 GB | Budget 24 GB VRAM |
-| **Q2_K** | ~10.7 GB | 2.0 | ~15 GB | Extreme budget, CPU offload |
-> **Recommendation**: Q4_K_M is the sweet spot for most users. Q8_0 if you have the VRAM and want minimal quantization error.
----
-## Usage (llama.cpp)
-```bash
-# Download a quant
-wget https://huggingface.co/GestaltLabs/Ornstein-3.6-27B-RYS-GGUF/resolve/main/Ornstein-3.6-27B-RYS-Q4_K_M.gguf
-# Run inference
-./llama-cli -m Ornstein-3.6-27B-RYS-Q4_K_M.gguf \
-    -p "Explain RYS layer duplication in one sentence." \
-    -n 512 --temp 0.6
-```
-### With llama-server (OpenAI-compatible API)
 ```bash
-./llama-server -m Ornstein-3.6-27B-RYS-Q4_K_M.gguf \
-    --host 0.0.0.0 --port 8080 \
-    -c 32768 --flash-attn
 ```
-### Recommended settings per quant
-| Quant | Context | Flash Attention | Offload layers |
-|-------|---------|-----------------|----------------|
-| Q8_0 | 4096+ | Yes | Full GPU |
-| Q6_K | 8192+ | Yes | Full GPU |
-| Q4_K_M | 16384+ | Yes | Full GPU |
-| Q3_K_M | 8192 | Yes | CPU offload if needed |
-| Q2_K | 4096 | Recommended | CPU + GPU mix |
----
-## Architecture Details
-- **Model**: Qwen 3.6 27B dense (hybrid attention)
-- **Layers**: 66 (64 original + 2 duplicated)
-- **Hidden size**: 5120
-- **Attention**: Gated Delta Net (linear) + full SDPA, full every 4 layers
-- **Heads**: 24 full / 4 KV / head_dim 256
-- **Context**: 262,144 tokens
-- **Vocab**: 248,320 tokens
-- **License**: Apache 2.0
----
-## Citation
-```bibtex
-@online{ng2026rys,
-  author = {Ng, David Noel},
-  title = {Retained-You-Seek (RYS)},
-  year = {2026},
-  url = {https://dnhkng.github.io/posts/rys/}
-}
 ```
-Base model: [GestaltLabs/Ornstein-3.6-27B](https://huggingface.co/GestaltLabs/Ornstein-3.6-27B)
-Base architecture: [unsloth/Qwen3.6-27B](https://huggingface.co/unsloth/Qwen3.6-27B)
----
 ## Support This Work
-I'm a PhD student in visual neuroscience at the University of Toronto who also happens to spend way too much time fine-tuning, merging, and quantizing open-weight models on rented H100s and a local DGX Spark. All training compute is self-funded — balancing GPU costs against a student budget. If my uploads have been useful to you, consider buying a PhD student a coffee. It goes a long way toward keeping these experiments running.
 **[Support on Ko-fi](https://ko-fi.com/djlougen)**
----
 ## License
-Apache 2.0 — inherited from Qwen 3.6 and Ornstein-3.6-27B.
-[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)

 # 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 GestaltLabs
+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.
+## ⚠️ Requires Patched llama.cpp
+RYS duplicates layer 33, which breaks the hardcoded attention-interval logic in stock llama.cpp. **Stock llama.cpp, Ollama, LM Studio, and similar tools will fail to load these GGUFs.**
+**Use this patched fork:** https://github.com/DJLougen/llama.cpp/tree/rys-qwen35
+The fork is fully backward-compatible with non-RYS Qwen3.5 models. It now also includes an **SSM tensor probing fallback**, so even legacy GGUFs load correctly without per-layer metadata.
+## Available Quantizations
+| File | Quant | Size | Notes |
+|------|-------|------|-------|
+| `ornstein-3.6-27b-rys-q8_0.gguf` | Q8_0 | ~27 GB | Near-lossless |
+| `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 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 |
+Sizes are approximate (exact depends on tokenizer overhead).
+## Quick Start
+### 1. 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 CPU-only.
+### 2. Download a GGUF
+Grab your preferred quant from the **Files** tab.
+### 3. Launch llama-server
+```bash
+./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
 ```
+* `--jinja` enables the Qwen3 thinking chat template.
+* `--flash-attn on` is recommended for long contexts.
+* `-ctk q4_0 -ctv q4_0` quantizes KV cache to 4-bit.
+## Note: Thinking Model
+This is a Qwen3-Thinking derivative. If you see raw `<think>...</think>` blocks appearing inline in responses, ensure `--jinja` is enabled. Recent llama.cpp builds default `--reasoning-format deepseek`, which splits reasoning into a separate `reasoning_content` field.
 ## 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)**
 ## License
+Apache 2.0