Instructions to use petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF",
	filename="ornith-1.0-35b-MTP-graft-down-Q4_0.gguf",
)

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

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0
# Run inference directly in the terminal:
llama cli -hf petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0
# Run inference directly in the terminal:
llama cli -hf petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0

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 petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0
# Run inference directly in the terminal:
./llama-cli -hf petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0

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 petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0
# Run inference directly in the terminal:
./build/bin/llama-cli -hf petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0

Use Docker

docker model run hf.co/petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0

LM Studio
Jan

vLLM

How to use petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-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": "petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0

Ollama
How to use petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF with Ollama:
```
ollama run hf.co/petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0
```

Unsloth Studio

How to use petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-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 petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-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 petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF to start chatting

How to use petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0

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": "petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0

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 petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Docker Model Runner
How to use petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF with Docker Model Runner:
```
docker model run hf.co/petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0
```

Lemonade

How to use petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF:Q4_0

Run and chat with the model

lemonade run user.Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF-Q4_0

List all available models

lemonade list

Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF / WINDOWS_ENDPOINT_QUICKSTART.md

petr567

docs: verify clean Windows download and endpoint recipe

38bb75c verified about 18 hours ago

preview code

Raw

History Blame Contribute Delete

7.44 kB

Windows quick start: Ornith MTP OpenAI endpoint

This recipe reproduces the fast Windows laptop profile measured for ornith-1.0-35b-MTP-graft-down-Q4_0.gguf:

Windows 11 with Docker Desktop using WSL2;
NVIDIA GPU visible to Docker (docker run --rm --gpus all ...);
8 GiB VRAM and approximately 44 GiB peak whole-system RAM in the measured RTX 4060 Laptop configuration;
llama.cpp b10066 (86a9c79f8);
dense and attention tensors on the GPU, MoE experts in host RAM;
integrated MTP plus n-gram speculative decoding;
OpenAI-compatible API on localhost only.

1. Prerequisites

Install or update:

NVIDIA Windows driver;
WSL2;
Docker Desktop with the WSL2 backend;
Git LFS or the Hugging Face CLI for the model download.

Confirm that Docker can see the GPU:

docker run --rm --gpus all nvidia/cuda:12.8.1-base-ubuntu24.04 nvidia-smi

The validated 16K profile needs roughly 43–45 GiB of total RAM in use on the test laptop. Close memory-heavy applications before starting on a 64 GiB machine. An 8 GiB GPU is sufficient because --cpu-moe keeps the large expert weights in host RAM.

2. Download the full MTP artifact

Create a model directory and an isolated virtual environment for the Hugging Face CLI. Do not upgrade huggingface_hub in a shared Python installation: new CLI releases can require a newer click than packages such as gTTS.

New-Item -ItemType Directory -Force C:\Models\Ornith-MTP | Out-Null

$HfVenv = 'C:\Models\Ornith-MTP\.hf-cli'
py -m venv $HfVenv
& "$HfVenv\Scripts\python.exe" -m pip install --upgrade pip
& "$HfVenv\Scripts\python.exe" -m pip install "huggingface_hub[hf_xet]==1.24.0"

& "$HfVenv\Scripts\hf.exe" download `
  petr567/Ornith-1.0-35B-MTP-Strix-Halo-Hybrid-GGUF `
  ornith-1.0-35b-MTP-graft-down-Q4_0.gguf `
  --local-dir C:\Models\Ornith-MTP

The download is approximately 19.4 GiB. Xet may spend time scanning chunks without continuously printing progress; leave the command running until the PowerShell prompt returns. A partially downloaded file is resumed on the next identical command.

This creates a real standalone copy downloaded from Hugging Face. The recipe does not use a hard link, symbolic link, or an already installed LM Studio model.

Expected file:

C:\Models\Ornith-MTP\ornith-1.0-35b-MTP-graft-down-Q4_0.gguf
Size: 20,329,342,112 bytes (18.933 GiB)
SHA-256: 365a7c02dfd320b9696f189d6dc12bd2b0eabb9f8e58ba9fc8cab3af93c0234b

This clean download and endpoint flow was re-tested on Windows on 19 July 2026, including checksum verification, Docker startup, /health, /v1/models, and /v1/chat/completions.

Verify it:

(Get-FileHash `
  C:\Models\Ornith-MTP\ornith-1.0-35b-MTP-graft-down-Q4_0.gguf `
  -Algorithm SHA256).Hash.ToLowerInvariant()

Do not use the separate LM Studio compatible, no integrated MTP file for this endpoint. It deliberately omits the speculative prediction layer.

3. Start the endpoint

The repository includes run-ornith-mtp-windows.ps1. Download it beside the model card or copy it from this repository, then run:

Set-ExecutionPolicy -Scope Process Bypass
.\run-ornith-mtp-windows.ps1 `
  -ModelDir C:\Models\Ornith-MTP `
  -Port 18081

The script pins the exact tested image:

ghcr.io/ggml-org/llama.cpp@sha256:1b3d1458ccda7287feab41b8001311acc03e24cde99ec0a2908fe83830562f38
llama.cpp b10066 (86a9c79f8)

Equivalent direct command:

$Image = 'ghcr.io/ggml-org/llama.cpp@sha256:1b3d1458ccda7287feab41b8001311acc03e24cde99ec0a2908fe83830562f38'
$ModelDir = 'C:\Models\Ornith-MTP'

docker run --detach --rm `
  --name ornith-mtp-api `
  --gpus all `
  --publish 127.0.0.1:18081:8080 `
  --mount "type=bind,source=$ModelDir,target=/models,readonly" `
  $Image `
  --model /models/ornith-1.0-35b-MTP-graft-down-Q4_0.gguf `
  --alias ornith-1.0-35b-mtp `
  --host 0.0.0.0 --port 8080 `
  --ctx-size 16384 --parallel 1 `
  --n-gpu-layers all --cpu-moe --no-mmap `
  --batch-size 2048 --ubatch-size 512 `
  --flash-attn on `
  --cache-type-k q8_0 --cache-type-v q8_0 `
  --jinja --metrics `
  --spec-type ngram-mod,draft-mtp `
  --spec-draft-n-max 2 `
  --spec-draft-n-min 1 `
  --spec-draft-p-min 0.20 `
  --spec-ngram-mod-n-min 48 `
  --spec-ngram-mod-n-max 64 `
  --spec-ngram-mod-n-match 24

The host publish address is deliberately 127.0.0.1, not 0.0.0.0. This prevents unauthenticated access from the LAN. Put nginx/Caddy with TLS and an API key in front of llama-server before exposing it to other machines.

4. Check readiness and API

Model loading can take tens of seconds:

docker logs -f ornith-mtp-api

In another terminal:

Invoke-RestMethod http://127.0.0.1:18081/health
Invoke-RestMethod http://127.0.0.1:18081/v1/models

OpenAI-compatible smoke request:

$Body = @{
  model = 'ornith-1.0-35b-mtp'
  messages = @(
    @{ role = 'user'; content = 'Reply with exactly: READY' }
  )
  temperature = 0
  max_tokens = 64
} | ConvertTo-Json -Depth 8

$Response = Invoke-RestMethod `
  -Method Post `
  -Uri http://127.0.0.1:18081/v1/chat/completions `
  -ContentType 'application/json' `
  -Body $Body

$Response.choices[0].message

Use these provider values in Orion or another OpenAI client:

Base URL: http://127.0.0.1:18081/v1
API key:  local-only-placeholder
Model:    ornith-1.0-35b-mtp

llama-server does not require the placeholder key in this localhost profile, but many OpenAI clients require a non-empty string in their configuration.

5. Observe and stop

# llama.cpp request and MTP statistics
Invoke-WebRequest http://127.0.0.1:18081/metrics -UseBasicParsing |
  Select-Object -ExpandProperty Content

# GPU telemetry
nvidia-smi `
  --query-gpu=temperature.gpu,utilization.gpu,memory.used,power.draw `
  --format=csv,noheader,nounits

# logs
docker logs --tail 100 ornith-mtp-api

# stop and release RAM/VRAM
docker stop ornith-mtp-api

Expected performance on the measured laptop

RTX 4060 Laptop 8 GiB, i7-13650HX and 64 GiB RAM:

Workload	Measured decode
1K prompt, ordinary output	28.74 tok/s
8K prompt, high MTP acceptance	66.88 tok/s
repeated 8K code	67.23 tok/s

The 60–67 tok/s range requires high speculative acceptance. Ordinary agent traffic should not be advertised as a fixed 67 tok/s minimum. The same hybrid without speculation measured about 27–30 tok/s.

Common problems

`could not select device driver ... gpu`

Update Docker Desktop and the NVIDIA driver, ensure Docker uses the WSL2 backend, restart Docker Desktop, then repeat the CUDA nvidia-smi check.

Container exits during model loading

Check docker logs ornith-mtp-api, close memory-heavy programs, and confirm at least about 45 GiB of RAM is available for the measured profile.

Only 1–2 tok/s

Do not replace --cpu-moe --n-gpu-layers all with an arbitrary partial layer split. The rejected 15-layer split forced expert traffic across CPU/GPU memory and reduced realistic prompt decode to about 1.2 tok/s.

MTP flags are unknown

The image is too old or a different runtime is being used. Run:

docker run --rm `
  ghcr.io/ggml-org/llama.cpp@sha256:1b3d1458ccda7287feab41b8001311acc03e24cde99ec0a2908fe83830562f38 `
  --version

It must report llama.cpp version 10066 (86a9c79f8) for this exact recipe.