Fix README: use ornstein3.6-27b.png image, move Ko-fi up, add YAML metadata

README.md

@@ -16,6 +16,8 @@ tags:
 - sovereign-ai
 ---
 
+[![Ornstein-3.6-27B-RYS](/GestaltLabs/Ornstein-3.6-27B-RYS-GGUF/resolve/main/ornstein3.6-27b.png)](/GestaltLabs/Ornstein-3.6-27B-RYS-GGUF/blob/main/ornstein3.6-27b.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.
@@ -24,29 +26,49 @@ GGUF quantizations of [GestaltLabs/Ornstein-3.6-27B-RYS](https://huggingface.co/
 
 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
+## 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.
 
-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.**
+**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).
 
-**Use this patched fork:** https://github.com/DJLougen/llama.cpp/tree/rys-qwen35
+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
 
-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.
+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 |
+| `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 balance |
+| `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 |
 
-Sizes are approximate (exact depends on tokenizer overhead).
+## 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
 
-## Quick Start
+## Usage
 
-### 1. Build the patched llama.cpp
+### Build the patched llama.cpp
 
 ```bash
 git clone https://github.com/DJLougen/llama.cpp.git
@@ -56,15 +78,15 @@ cmake -B build -DGGML_CUDA=ON -DCMAKE_BUILD_TYPE=Release
 cmake --build build -j
 ```
 
-Drop `-DGGML_CUDA=ON` for CPU-only.
+Drop `-DGGML_CUDA=ON` for a CPU-only build. The patch touches the GGUF loader and three model forward files; backend selection is independent.
 
-### 2. Download a GGUF
-
-Grab your preferred quant from the **Files** tab.
-
-### 3. Launch llama-server
+### 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 \
@@ -73,20 +95,6 @@ Grab your preferred quant from the **Files** tab.
     -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