Instructions to use FoolDev/Thanatos-27B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use FoolDev/Thanatos-27B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="FoolDev/Thanatos-27B")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("FoolDev/Thanatos-27B", dtype="auto")

llama-cpp-python

How to use FoolDev/Thanatos-27B with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="FoolDev/Thanatos-27B",
	filename="Thanatos-27B.Q4_K_M.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": [
				{
					"type": "text",
					"text": "Describe this image in one sentence."
				},
				{
					"type": "image_url",
					"image_url": {
						"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
					}
				}
			]
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use FoolDev/Thanatos-27B with llama.cpp:

Install from brew

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

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf FoolDev/Thanatos-27B:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf FoolDev/Thanatos-27B: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 FoolDev/Thanatos-27B:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf FoolDev/Thanatos-27B: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 FoolDev/Thanatos-27B:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf FoolDev/Thanatos-27B:Q4_K_M

Use Docker

docker model run hf.co/FoolDev/Thanatos-27B:Q4_K_M

LM Studio
Jan

vLLM

How to use FoolDev/Thanatos-27B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "FoolDev/Thanatos-27B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "FoolDev/Thanatos-27B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/FoolDev/Thanatos-27B:Q4_K_M

SGLang

How to use FoolDev/Thanatos-27B with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "FoolDev/Thanatos-27B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "FoolDev/Thanatos-27B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "FoolDev/Thanatos-27B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "FoolDev/Thanatos-27B",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Ollama
How to use FoolDev/Thanatos-27B with Ollama:
```
ollama run hf.co/FoolDev/Thanatos-27B:Q4_K_M
```

Unsloth Studio new

How to use FoolDev/Thanatos-27B 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 FoolDev/Thanatos-27B 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 FoolDev/Thanatos-27B to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for FoolDev/Thanatos-27B to start chatting

Pi new

How to use FoolDev/Thanatos-27B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf FoolDev/Thanatos-27B: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": "FoolDev/Thanatos-27B:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use FoolDev/Thanatos-27B with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf FoolDev/Thanatos-27B: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 FoolDev/Thanatos-27B:Q4_K_M

Run Hermes

hermes

Docker Model Runner
How to use FoolDev/Thanatos-27B with Docker Model Runner:
```
docker model run hf.co/FoolDev/Thanatos-27B:Q4_K_M
```

Lemonade

How to use FoolDev/Thanatos-27B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull FoolDev/Thanatos-27B:Q4_K_M

Run and chat with the model

lemonade run user.Thanatos-27B-Q4_K_M

List all available models

lemonade list

FoolDev commited on 10 days ago

Commit

e9c7366

1 Parent(s): 07fa120

docs: align README + CHANGELOG + Modelfile + examples with the qwen36 re-stamp (07fa120)

Browse files

Files changed (4) hide show

CHANGELOG.md +34 -36
Modelfile +20 -10
README.md +87 -58
examples/README.md +13 -11

CHANGELOG.md CHANGED Viewed

@@ -8,45 +8,43 @@ and documentation**, not the underlying base model.
 ## [Unreleased]
 ### Changed
-- **Bundle re-stamped `general.architecture: 'qwen36'` → `'qwen35'`**
-  in commit 964e418 (HF-direct `hf upload`). Reverses the v0.6.0
-  `qwen36` stamp from e1f78fa, which itself flipped qwen35 → qwen36
-  deliberately. The proximate trigger: today's subagent recheck
-  found no PR or tracking issue for a `qwen36` arch entry in either
-  `ggml-org/llama.cpp` or `ollama/ollama`, and the actual Qwen 3.6
-  family upstream already loads under `qwen35moe` — so the qwen36
-  stamp was going to stay ahead-of-upstream indefinitely, not for a
-  release or two. Day-to-day cost (every fresh
-  `ollama run hf.co/FoolDev/Thanatos-27B` hitting a 500, every user
-  needing `make heal-hf` to recover) outweighed the
-  architectural-honesty benefit. Tensor data stays byte-identical;
-  only the `general.architecture` KV (and the namespaced KV keys)
-  flips. New bundle blob SHA is
-  `5ed60d0af4650a854b1755bd392f9aef4872643dc25a254bc68043fa638392a0`
-  — the same hash `make load-bundle` and `make heal-hf` have been
-  producing locally this whole session.
-- README "Heads up" callout (qwen36 load-failure warning) removed.
-  TL;DR collapsed back from "three paths around it" to a single
-  one-liner: `ollama run hf.co/FoolDev/Thanatos-27B` now loads on
-  stock Ollama. `make heal-hf` is mentioned as the recovery path
-  for anyone who pulled the v0.6.0 bundle and still has the qwen36
-  blob in their store.
-- README "Architecture" section, Quick start option A,
-  `examples/README.md` Ollama setup, and `Modelfile` preamble all
-  collapsed back to the pre-qwen36 single-path narrative. The
-  long "qwen36 is unloadable / here are the workarounds" prose
-  was replaced with a one-paragraph history note pointing at
-  commit 964e418 + the [CHANGELOG](CHANGELOG.md) entry. Net -39
-  lines across the three doc surfaces. `scripts/rename_arch.py`
-  example in Architecture flipped from qwen36 → qwen35 to qwen35
-  → qwen36 to match its new "tool kept around for future flips"
-  framing.
 ### Note
 - `make heal-hf`, `make load-bundle`, and `scripts/rename_arch.py`
-  stay in the repo and stay useful. heal-hf is idempotent on qwen35
-  (skips early); load-bundle does a no-op rebadge on a qwen35 bundle
-  and runs `ollama create` as normal.
 ### Added
 - `scripts/check.sh` now greps for the `VAR="$(cmd 2>/dev/null | filter)"`

 ## [Unreleased]
 ### Changed
+- **Bundle re-stamped `general.architecture: 'qwen35'` → `'qwen36'`**
+  in commit `07fa120` (HF-direct `hf upload`). Re-reverses 964e418,
+  returning to the v0.6.0 stamp choice from `e1f78fa`. Tensor data
+  is byte-identical across all three flips; only the
+  `general.architecture` KV (and the namespaced KV keys) moves.
+  The new blob SHA
+  `b2b63b941d714b0aff5e9afc9b337e7607b84e371ba991c182d92d0f7805c0e9`
+  is **bit-identical to the original v0.6.0 qwen36 bundle** —
+  content-addressed return to a known artifact, so anyone who pulled
+  v0.6.0 still has a usable cached blob; `make heal-hf` will skip
+  the re-fetch if their old blob is still in the store.
+  The flip restores the architecturally-honest label even though
+  upstream `qwen36` arch support is still not present in
+  `ggml-org/llama.cpp` or `ollama/ollama` as of 2026-05-19 (verified
+  by recheck; no open PR adds it). Day-to-day cost of the qwen36
+  stamp is back: a fresh `ollama run hf.co/FoolDev/Thanatos-27B`
+  pulls 17 GB and 500s with `unable to load model: <blob>`,
+  requiring one `make heal-hf` to rebadge in place before the same
+  tag loads. The trade-off here favors honesty + matching the
+  upstream Qwen 3.6 family identity over no-rebadge ergonomics
+  on the HF-bridge path.
+- README "Heads up" callout, three-path TL;DR, Architecture
+  workaround prose, Modelfile preamble, and `examples/README.md`
+  Ollama setup all restored to the pre-collapse form (i.e. the
+  v0.6.0-era doc surface, updated for the new SHA and the third
+  flip). `scripts/rename_arch.py` example direction in
+  Architecture flipped back to qwen36 → qwen35 — the rebadge
+  direction users actually run to make the bundle loadable, not
+  the future-honesty direction.
 ### Note
 - `make heal-hf`, `make load-bundle`, and `scripts/rename_arch.py`
+  are now back to load-bearing rather than dormant.
+  `make heal-hf` is the canonical recovery path; `make load-bundle`
+  builds a separate local tag without touching the HF-bridge
+  manifest; `make build` bypasses the bundle entirely (downloads
+  qwen35-stamped from unsloth).
 ### Added
 - `scripts/check.sh` now greps for the `VAR="$(cmd 2>/dev/null | filter)"`

Modelfile CHANGED Viewed

@@ -7,18 +7,28 @@
 # image input, or wait for the fix. See the Vision section in README.md.
 #
 # This repo bundles a single GGUF: Thanatos-27B.Q4_K_M.gguf (~17 GB),
-# stamped `general.architecture: 'qwen35'` — the arch entry every
-# released llama.cpp / Ollama loads under for the Qwen 3.6 hybrid
-# SSM + attention family. `ollama create thanatos-27b -f Modelfile &&
-# ollama run thanatos-27b` loads it directly. See README
-# "Architecture" for the qwen35 -> qwen36 -> qwen35 history through
-# v0.6.0.
 #
 # For other quants (Q3_K_S, Q5_K_M, Q6_K, etc.), `make build QUANT=Q3_K_S`
-# downloads the chosen quant from unsloth/Qwen3.6-27B-GGUF and patches
-# FROM in a temp Modelfile copy. The Q3_K_S used to ship in this repo;
-# it was removed so HF's Ollama bridge picks Q4_K_M as the default
-# `:latest` tag instead of Q3_K_S (alphabetically-first heuristic).
 #
 # Other GGUF sources (use with `make build GGUF_PATH=...`):
 #     https://huggingface.co/unsloth/Qwen3.6-27B-GGUF

 # image input, or wait for the fix. See the Vision section in README.md.
 #
 # This repo bundles a single GGUF: Thanatos-27B.Q4_K_M.gguf (~17 GB),
+# stamped `general.architecture: 'qwen36'` — the architecturally-honest
+# label for the Qwen 3.6 family. No released llama.cpp / Ollama
+# recognizes that string yet; both register the stack as
+# `qwen35` / `qwen35moe`. A direct `ollama create -f Modelfile`
+# against the bundle therefore fails with
+# `unknown model architecture: 'qwen36'`. Two recovery paths:
+#
+#   make load-bundle    # one-shot: smudge LFS pointer, rebadge a local
+#                       # copy qwen36 -> qwen35, run `ollama create`.
+#   make heal-hf        # rebadge an already-pulled `hf.co/...` tag's
+#                       # blob in place (qwen36 -> qwen35) and rewrite
+#                       # the manifest digest.
+#
+# See README "Architecture" for the qwen35 -> qwen36 -> qwen35 ->
+# qwen36 stamp history through commit 07fa120.
 #
 # For other quants (Q3_K_S, Q5_K_M, Q6_K, etc.), `make build QUANT=Q3_K_S`
+# downloads the chosen quant from unsloth/Qwen3.6-27B-GGUF (qwen35-stamped,
+# loads on stock loaders, no rebadge needed) and patches FROM in a temp
+# Modelfile copy. The Q3_K_S used to ship in this repo; it was removed so
+# HF's Ollama bridge picks Q4_K_M as the default `:latest` tag instead of
+# Q3_K_S (alphabetically-first heuristic).
 #
 # Other GGUF sources (use with `make build GGUF_PATH=...`):
 #     https://huggingface.co/unsloth/Qwen3.6-27B-GGUF

README.md CHANGED Viewed

@@ -61,33 +61,51 @@ pipeline_tag: image-text-to-text
 A personal sibling to [`FoolDev/Janus-35B`](https://huggingface.co/FoolDev/Janus-35B). Same teacher (Claude Opus 4.7), same dataset family, but built on the **dense** [Qwen/Qwen3.6-27B](https://huggingface.co/Qwen/Qwen3.6-27B) base instead of the 35B-A3B MoE. Smaller, easier to deploy, no expert-routing surprises.
 ## TL;DR
-One-liner via Hugging Face (pulls a GGUF + this repo's root-level
-`template` / `system` / `params` files, including the tool-calling
-template — HF's Ollama bridge ingests those three files, not
-`Modelfile`):
 ```bash
-ollama run hf.co/FoolDev/Thanatos-27B           # ~17 GB Q4_K_M, qwen35-stamped, loads on stock Ollama
-```
-If you pulled the bundle while it was qwen36-stamped (v0.6.0
-through the early hours of 2026-05-19) the load will still 500 on
-that stale blob — `make heal-hf` rebadges it in place. Fresh
-pulls after this fix go straight through.
-For other quants (Q3_K_S ~12 GB, Q5_K_M ~20 GB, etc.):
-```bash
 git clone https://huggingface.co/FoolDev/Thanatos-27B && cd Thanatos-27B
-make build QUANT=Q3_K_S                  # downloads from unsloth/Qwen3.6-27B-GGUF
 ollama run thanatos-27b
 ```
-For other quants (Q3_K_S ~12 GB, Q5_K_M ~20 GB, etc.), `make build
-QUANT=Q3_K_S` is the simplest path. See [Quick start](#quick-start)
-below for the full matrix.
 For image input use llama.cpp directly — Ollama vision is broken for
 this architecture upstream (see [Vision](#vision)).
@@ -139,7 +157,7 @@ This repo ships a single GGUF to back the HF/Ollama "Use this model"
 widget — `Thanatos-27B.Q4_K_M.gguf` (~17 GB, qwen36-stamped):
 ```bash
-ollama run hf.co/FoolDev/Thanatos-27B           # 17 GB Q4_K_M, qwen36 — fails today, see Heads-up above
 ```
 For 16 GB GPUs / unified-memory laptops — and as a working-today
@@ -176,43 +194,46 @@ If you want the safetensors for `transformers`, fetch them from [`Qwen/Qwen3.6-2
   current loader compatibility.
 - Multi-token prediction (MTP) head trained for speculative decoding
-**The bundled GGUF declares `general.architecture: 'qwen35'`** — the
-arch entry every released llama.cpp / Ollama loads the Qwen 3.6
-hybrid SSM + attention family under. Upstream `ggml-org/llama.cpp`
-and `ollama/ollama` register the stack as `qwen35` / `qwen35moe`,
-including for actual Qwen 3.6-family weights (the upstream
-`Qwen3.6-35B-A3B_Q8_0` GGUFs load under `qwen35moe`); no `qwen36`
-arch entry exists, and there's no open PR or tracking issue for
-one — `qwen35` is the canonical load-time label.
-`ollama run hf.co/FoolDev/Thanatos-27B` and `llama-server -m
-Thanatos-27B.Q4_K_M.gguf` both load directly on current stock
-loaders.
-**History.** Through v0.6.0 the bundle was stamped `qwen36` — the
-architecturally-honest label, but unloadable on stock loaders.
-Users needed `make load-bundle` or `make heal-hf` to rebadge
-before loading. The stamp was reverted to `qwen35` in commit
-`964e418` to stop forcing the workaround on every fresh pull;
-tensor data was byte-identical across both stamps, only the
-`general.architecture` KV (and namespaced KV keys) flipped. See
-the [CHANGELOG](CHANGELOG.md) entry for the full rationale.
-`scripts/rename_arch.py` is kept in the repo as a generic GGUF
-arch renamer (metadata only, tensors byte-identical) for future
-flips:
 ```bash
 python3 scripts/rename_arch.py \
-    --from-arch qwen35 --to-arch qwen36 \
     Thanatos-27B.Q4_K_M.gguf \
-    Thanatos-27B.Q4_K_M.qwen36.gguf
 ```
-If you pulled the bundle while it was still qwen36-stamped (v0.6.0
-through 2026-05-19) and have the broken blob in your local Ollama
-store, `make heal-hf` rebadges it in place so the same
-`hf.co/...` tag becomes loadable without re-pulling.
 ## Quick start
@@ -221,18 +242,25 @@ store, `make heal-hf` rebadges it in place so the same
 Three paths:
 ```bash
-# A. Pull straight from HF (gets the bundled Q4_K_M GGUF + the
-#    root-level template / system / params files in one step):
-ollama run hf.co/FoolDev/Thanatos-27B           # 17 GB Q4_K_M, qwen35-stamped
 # B. Build a local `thanatos-27b` tag from THIS repo's bundle
-#    (LFS smudge if needed, then `ollama create`). Useful if you
-#    want a bare local tag rather than the `hf.co/...` path:
 make load-bundle                                 # creates local tag thanatos-27b
 ollama run thanatos-27b
 # C. Bypass the bundle: download a qwen35-stamped GGUF from unsloth
-#    and build locally. Loads on every current llama.cpp / Ollama.
 make build                                              # Q4_K_M  -> thanatos-27b
 make build QUANT=Q3_K_S                                 # 12 GB smaller quant
 make build QUANT=Q5_K_M                                 # 20 GB higher quality
@@ -260,14 +288,15 @@ python examples/ollama_chat.py      # full demo: chat, streaming, tools, OpenAI-
 The bundled `Thanatos-27B.Q4_K_M.gguf` is `qwen36`-stamped — every row
 below assumes you've rebadged it to `qwen35` per
-[Architecture](#architecture), or that you're pulling a `qwen35`-stamped
-GGUF from `unsloth/Qwen3.6-27B-GGUF` instead. The "fails today with
 `unknown model architecture: 'qwen36'`" caveat applies to every row
 until that's done.
 | App | How to load this model |
 |---|---|
-| **Ollama** | `ollama run hf.co/FoolDev/Thanatos-27B` (default Q4_K_M). Pulls the GGUF + the root-level `template` / `system` / `params` files in one step (HF's Ollama bridge ingests these three files; it does **not** read `Modelfile`). For other quants, or to bypass the qwen36 block today, `make build QUANT=Q3_K_S` downloads from unsloth (qwen35-stamped) and creates a local Ollama tag using the `Modelfile`, which is kept in sync with the bridge files. |
 | **LM Studio** | Search → `FoolDev/Thanatos-27B` → pick `Thanatos-27B.Q4_K_M.gguf`. Uses the GGUF's embedded jinja chat template (Qwen 3.6 ChatML); set the system prompt manually from the `SYSTEM` block in this repo's `Modelfile`. |
 | **Jan** | Hub → "Import from Hugging Face" → `FoolDev/Thanatos-27B`. Same template behavior as LM Studio. |
 | **llama.cpp** | `hf download FoolDev/Thanatos-27B Thanatos-27B.Q4_K_M.gguf --local-dir .` then `llama-server -m Thanatos-27B.Q4_K_M.gguf` (or `llama-cli`, `llama-mtmd-cli` for vision via the upstream `mmproj-F16.gguf`). |

 A personal sibling to [`FoolDev/Janus-35B`](https://huggingface.co/FoolDev/Janus-35B). Same teacher (Claude Opus 4.7), same dataset family, but built on the **dense** [Qwen/Qwen3.6-27B](https://huggingface.co/Qwen/Qwen3.6-27B) base instead of the 35B-A3B MoE. Smaller, easier to deploy, no expert-routing surprises.
+> **Heads up** — the bundled GGUF is stamped
+> `general.architecture: 'qwen36'`, the architecturally-honest
+> label for the Qwen 3.6 family. **No released llama.cpp / Ollama
+> recognizes it yet** (`ggml-org/llama.cpp` and `ollama/ollama`
+> register the hybrid SSM + attention stack as `qwen35` /
+> `qwen35moe`; verified 2026-05-19). A fresh
+> `ollama run hf.co/FoolDev/Thanatos-27B` pulls ~17 GB and then
+> 500s with `unable to load model: <blob>`. Run `make heal-hf`
+> once after the failure to rebadge the cached blob qwen36 →
+> qwen35 in place — subsequent runs load. Or skip the bundle
+> entirely with `make build` (qwen35-stamped from unsloth, loads
+> on stock Ollama). See [Architecture](#architecture) for the
+> stamp history and trade-off.
 ## TL;DR
+Three paths, in increasing order of disk re-use:
 ```bash
+# A. Pull from HF (gets the bundled qwen36 GGUF + root-level
+#    template / system / params files). First `ollama run` 500s
+#    on the qwen36 blob; `make heal-hf` rebadges it in place.
+ollama run hf.co/FoolDev/Thanatos-27B           # 17 GB Q4_K_M, qwen36
+git clone https://huggingface.co/FoolDev/Thanatos-27B && cd Thanatos-27B
+make heal-hf                                    # rebadge cached blob -> qwen35
+ollama run hf.co/FoolDev/Thanatos-27B           # now loads
+# B. Build a local `thanatos-27b` tag from THIS repo's bundle
+#    (smudges the LFS pointer via `hf download`, rebadges
+#    qwen36 -> qwen35 in a local copy, runs `ollama create`):
+git clone https://huggingface.co/FoolDev/Thanatos-27B && cd Thanatos-27B
+make load-bundle                                # creates local tag thanatos-27b
+ollama run thanatos-27b
+# C. Bypass the bundle: download a qwen35-stamped GGUF from
+#    unsloth/Qwen3.6-27B-GGUF and build locally — no rebadge
+#    needed. Loads on every current llama.cpp / Ollama.
 git clone https://huggingface.co/FoolDev/Thanatos-27B && cd Thanatos-27B
+make build                                      # Q4_K_M -> thanatos-27b
 ollama run thanatos-27b
 ```
+For other quants (Q3_K_S ~12 GB, Q5_K_M ~20 GB, etc.) path **C**
+is simplest: `make build QUANT=Q3_K_S`. See
+[Quick start](#quick-start) below for the full matrix.
 For image input use llama.cpp directly — Ollama vision is broken for
 this architecture upstream (see [Vision](#vision)).
 widget — `Thanatos-27B.Q4_K_M.gguf` (~17 GB, qwen36-stamped):
 ```bash
+ollama run hf.co/FoolDev/Thanatos-27B           # 17 GB Q4_K_M, qwen36 — fails today, see Heads up above
 ```
 For 16 GB GPUs / unified-memory laptops — and as a working-today
   current loader compatibility.
 - Multi-token prediction (MTP) head trained for speculative decoding
+**The bundled GGUF declares `general.architecture: 'qwen36'`** — the
+architecturally-honest label for the Qwen 3.6 family. The cost is
+that no released `ggml-org/llama.cpp` or `ollama/ollama` registers
+that string yet: both register the hybrid SSM + attention stack as
+`qwen35` / `qwen35moe`, including for actual Qwen 3.6-family weights
+(the upstream `Qwen3.6-35B-A3B_Q8_0` GGUFs load under `qwen35moe`).
+No `qwen36` arch entry exists in either repo, and as of 2026-05-19
+there's still no open PR or tracking issue adding one. A
+`qwen36`-stamped GGUF therefore fails to load on stock loaders with
+`unknown model architecture: 'qwen36'` until the metadata is
+rebadged.
+`scripts/rename_arch.py` does the rebadge: a metadata-only flip
+that leaves all tensor bytes intact. `make heal-hf` automates it
+for an already-pulled `hf.co/FoolDev/Thanatos-27B` tag (rewrites
+the blob + the manifest's model-layer digest, idempotent on
+already-rebadged tags); `make load-bundle` builds a separate
+local `thanatos-27b` tag from a rebadged copy without touching the
+HF-bridge manifest. Both produce a tag that loads on stock Ollama:
 ```bash
 python3 scripts/rename_arch.py \
+    --from-arch qwen36 --to-arch qwen35 \
     Thanatos-27B.Q4_K_M.gguf \
+    Thanatos-27B.Q4_K_M.qwen35.gguf
 ```
+**History.** The bundle has flipped stamp three times. v0.6.0
+shipped `qwen36` (commit `e1f78fa`, architecturally-honest).
+Commit `964e418` (2026-05-19) reverted to `qwen35` to stop forcing
+the rebadge workaround on every fresh pull. Commit `07fa120`
+(same day) flipped it back to `qwen36` — restoring the honest
+label and accepting that stock Ollama users need one `make
+heal-hf` after the first failed run. Tensor data is byte-identical
+across all three states; only the `general.architecture` KV (and
+namespaced KV keys) flips. The current bundle blob SHA is
+`b2b63b941d714b0aff5e9afc9b337e7607b84e371ba991c182d92d0f7805c0e9`,
+identical to the original v0.6.0 qwen36 blob — content-addressed
+return to a known artifact. See the
+[CHANGELOG](CHANGELOG.md) entry for the trade-off rationale.
 ## Quick start
 Three paths:
 ```bash
+# A. Pull straight from HF (gets the bundled qwen36 Q4_K_M GGUF +
+#    the root-level template / system / params files in one step).
+#    The first `ollama run` will 500 on the qwen36 blob —
+#    `make heal-hf` rebadges it in place so the same tag loads.
+ollama run hf.co/FoolDev/Thanatos-27B           # 17 GB pull, then 500
+make heal-hf                                    # rebadge cached blob in store
+ollama run hf.co/FoolDev/Thanatos-27B           # now loads
 # B. Build a local `thanatos-27b` tag from THIS repo's bundle
+#    (smudges LFS pointer via `hf download` if needed, rebadges
+#    qwen36 -> qwen35 in a local copy, runs `ollama create`).
+#    Useful if you want a bare local tag rather than the
+#    hf.co/... path:
 make load-bundle                                 # creates local tag thanatos-27b
 ollama run thanatos-27b
 # C. Bypass the bundle: download a qwen35-stamped GGUF from unsloth
+#    and build locally. Loads on every current llama.cpp / Ollama
+#    with no rebadge step.
 make build                                              # Q4_K_M  -> thanatos-27b
 make build QUANT=Q3_K_S                                 # 12 GB smaller quant
 make build QUANT=Q5_K_M                                 # 20 GB higher quality
 The bundled `Thanatos-27B.Q4_K_M.gguf` is `qwen36`-stamped — every row
 below assumes you've rebadged it to `qwen35` per
+[Architecture](#architecture) (via `make heal-hf` or `make load-bundle`),
+or that you're pulling a `qwen35`-stamped GGUF from
+`unsloth/Qwen3.6-27B-GGUF` instead. The "fails today with
 `unknown model architecture: 'qwen36'`" caveat applies to every row
 until that's done.
 | App | How to load this model |
 |---|---|
+| **Ollama** | `ollama run hf.co/FoolDev/Thanatos-27B` (default Q4_K_M) pulls the GGUF + the root-level `template` / `system` / `params` files in one step (HF's Ollama bridge ingests these three files; it does **not** read `Modelfile`). The first `ollama run` 500s on the qwen36 blob — run `make heal-hf` once to rebadge it in place, after which the same tag loads. To skip the rebadge entirely, `make build QUANT=Q4_K_M` (or another quant) downloads a qwen35-stamped GGUF from unsloth and creates a local Ollama tag using the `Modelfile`, which is kept in sync with the bridge files. |
 | **LM Studio** | Search → `FoolDev/Thanatos-27B` → pick `Thanatos-27B.Q4_K_M.gguf`. Uses the GGUF's embedded jinja chat template (Qwen 3.6 ChatML); set the system prompt manually from the `SYSTEM` block in this repo's `Modelfile`. |
 | **Jan** | Hub → "Import from Hugging Face" → `FoolDev/Thanatos-27B`. Same template behavior as LM Studio. |
 | **llama.cpp** | `hf download FoolDev/Thanatos-27B Thanatos-27B.Q4_K_M.gguf --local-dir .` then `llama-server -m Thanatos-27B.Q4_K_M.gguf` (or `llama-cli`, `llama-mtmd-cli` for vision via the upstream `mmproj-F16.gguf`). |

examples/README.md CHANGED Viewed

@@ -19,24 +19,26 @@ from the `Modelfile` / bridge files.
 ### Ollama
-Pull straight from HF (gets the bundled Q4_K_M GGUF + this repo's
-root-level `template` / `system` / `params` files via HF's Ollama
-bridge):
 ```bash
 ollama pull hf.co/FoolDev/Thanatos-27B           # 17 GB Q4_K_M (only bundled quant)
 pip install requests
 MODEL=hf.co/FoolDev/Thanatos-27B python ollama_chat.py
 ```
-If you pulled before commit `964e418` (the qwen35 re-stamp) and
-still have the broken qwen36 blob in your Ollama store, run
-`cd .. && make heal-hf` once to rebadge it in place. Fresh pulls
-after the re-stamp go straight through.
-For a non-bundled quant (e.g. Q3_K_S ~12 GB, Q5_K_M ~20 GB),
-`make build QUANT=...` downloads from `unsloth/Qwen3.6-27B-GGUF`
-and creates a local `thanatos-27b` tag:
 ```bash
 cd ..  &&  make build QUANT=Q3_K_S  &&  cd examples

 ### Ollama
+The bundle is stamped `general.architecture: 'qwen36'` and no
+released llama.cpp / Ollama recognizes that string yet (both
+register the Qwen 3.6 stack as `qwen35` / `qwen35moe` — see
+[../README.md#architecture](../README.md#architecture)). The first
+`ollama run` against the HF-bridge tag pulls 17 GB and then 500s
+with `unable to load model: <blob>`. Run `make heal-hf` once
+after the failure to rebadge the cached blob in place — the same
+tag loads on subsequent runs:
 ```bash
 ollama pull hf.co/FoolDev/Thanatos-27B           # 17 GB Q4_K_M (only bundled quant)
+cd ..  &&  make heal-hf  &&  cd examples         # rebadge cached blob qwen36 -> qwen35
 pip install requests
 MODEL=hf.co/FoolDev/Thanatos-27B python ollama_chat.py
 ```
+For a non-bundled quant (e.g. Q3_K_S ~12 GB, Q5_K_M ~20 GB) — or
+to skip the rebadge entirely — `make build QUANT=...` downloads
+a qwen35-stamped GGUF from `unsloth/Qwen3.6-27B-GGUF` and creates
+a local `thanatos-27b` tag:
 ```bash
 cd ..  &&  make build QUANT=Q3_K_S  &&  cd examples