Instructions to use olluyeancah/margin-pilot1-gguf with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use olluyeancah/margin-pilot1-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 olluyeancah/margin-pilot1-gguf:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf olluyeancah/margin-pilot1-gguf:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf olluyeancah/margin-pilot1-gguf:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf olluyeancah/margin-pilot1-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 olluyeancah/margin-pilot1-gguf:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf olluyeancah/margin-pilot1-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 olluyeancah/margin-pilot1-gguf:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf olluyeancah/margin-pilot1-gguf:Q4_K_M

Use Docker

docker model run hf.co/olluyeancah/margin-pilot1-gguf:Q4_K_M

LM Studio
Jan
Ollama
How to use olluyeancah/margin-pilot1-gguf with Ollama:
```
ollama run hf.co/olluyeancah/margin-pilot1-gguf:Q4_K_M
```
Unsloth Desktop

How to use olluyeancah/margin-pilot1-gguf with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf olluyeancah/margin-pilot1-gguf:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "olluyeancah/margin-pilot1-gguf:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use olluyeancah/margin-pilot1-gguf with Docker Model Runner:
```
docker model run hf.co/olluyeancah/margin-pilot1-gguf:Q4_K_M
```

Lemonade

How to use olluyeancah/margin-pilot1-gguf with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull olluyeancah/margin-pilot1-gguf:Q4_K_M

Run and chat with the model

lemonade run user.margin-pilot1-gguf-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use olluyeancah/margin-pilot1-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 olluyeancah/margin-pilot1-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 olluyeancah/margin-pilot1-gguf:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use olluyeancah/margin-pilot1-gguf with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf olluyeancah/margin-pilot1-gguf:Q4_K_M

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 "olluyeancah/margin-pilot1-gguf:Q4_K_M" \
  --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"

Margin Pilot 1 — Q4_K_M

An experimental merged SFT-LoRA model for MeetingIntro Margin, derived from empero-ai/Qwen3.8-4B-Distill revision c83cb7aa2999d2f35c43e9ae0634a30eb8985a1e.

This package is Pilot 1, not the later DPO adapter. It is a community Empero/Qwen3.5-architecture derivative, not an official Qwen release.

Evaluation limits

The original developer suite reported 279/288 passing model requests, with three repetitions per fixture. Known failures remain. A separate thinking-disabled workstation subset passed 42/60 using evaluator v3. Those are different evaluation conditions and are not interchangeable. Native validation on a 24 GB Mac passed 76/78 cases; the two misses were an unnecessary tentative-RSVP refusal and a date move missing its destination. The eight-holiday all-day batch passed. All 45 roadmap guards and 22 managed-setup guards passed. A clean managed installation using the published, pinned model succeeded on 2026-09-13 UTC. Median native latency was 4.04 seconds, p95 15.77 seconds, with concurrent build activity; this is not a 16 GB hardware qualification. No claim of 99% real-world reliability is made.

Tools must enforce granted roots, preview and confirmation, reversible operations, and independent memory provenance checks. This model does not confer permission to execute its output.

Reproducibility

Quantized size: 2,783,446,336 bytes. SHA-256 is in SHA256SUMS. Use Ollama with an 8192-token context and temperature 0. Runtime and thinking mode materially affect results; qualify the exact app/runtime combination before deployment.

The model contains merged adapter weights. Training data, personal files, credentials, and user conversations are not included in this distribution package. No guarantee is made that a trained model cannot reproduce training examples.

Changes and attribution

Upstream model by Empero AI, based on Qwen3.5 architecture. Modified with Margin-specific SFT LoRA training, merged to canonical tensor names, and quantized to Q4_K_M. Auxiliary MTP tensors preserved from the pinned source. Licensed under Apache 2.0; see LICENSE and upstream attribution.