Instructions to use FINAL-Bench/POCKET-Darwin-180B-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 FINAL-Bench/POCKET-Darwin-180B-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 FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL

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 FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./llama-cli -hf FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL

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 FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./build/bin/llama-cli -hf FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL

Use Docker

docker model run hf.co/FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL

LM Studio
Jan

vLLM

How to use FINAL-Bench/POCKET-Darwin-180B-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "FINAL-Bench/POCKET-Darwin-180B-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": "FINAL-Bench/POCKET-Darwin-180B-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL

Ollama
How to use FINAL-Bench/POCKET-Darwin-180B-GGUF with Ollama:
```
ollama run hf.co/FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL
```
Unsloth Desktop

How to use FINAL-Bench/POCKET-Darwin-180B-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL

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": "FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use FINAL-Bench/POCKET-Darwin-180B-GGUF with Docker Model Runner:
```
docker model run hf.co/FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL
```

Lemonade

How to use FINAL-Bench/POCKET-Darwin-180B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL

Run and chat with the model

lemonade run user.POCKET-Darwin-180B-GGUF-UD-Q4_K_XL

List all available models

lemonade list

Hermes Agent

How to use FINAL-Bench/POCKET-Darwin-180B-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 FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL

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 FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use FINAL-Bench/POCKET-Darwin-180B-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL

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 "FINAL-Bench/POCKET-Darwin-180B-GGUF:UD-Q4_K_XL" \
  --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"

POCKET-Darwin-180B: a clean R0 vs R3 test on held-out SuperGPQA (+2.45 points)

by SeaWolf-AI - opened about 16 hours ago

Discussion

SeaWolf-AI

FINAL_Bench org about 16 hours ago

We released POCKET-Darwin-180B, a 4-bit GGUF build of Darwin-180B-RSI (R3) that runs on a laptop or a CPU-only server.
https://huggingface.co/FINAL-Bench/POCKET-Darwin-180B-GGUF

How it was built makes it a useful test of RSI itself. We took the public Unsloth UD-Q4_K_XL build of the parent model (Qwen3.8-Flash-Next) and replaced only the 300 tensors that our self-improvement training changed. Everything else is byte-identical. So comparing POCKET with the parent 4-bit build isolates the effect of RSI: same quantized experts, same everything, except the 2.78% of bytes RSI touched.

Held-out SuperGPQA, 1,000 questions never used in training or selection, 4 samples each, 16K generation budget:

Parent 4-bit: 59.10%
POCKET (R3): 61.55%
Paired difference: +2.45 points, 95% CI [+1.54, +3.36]

Where the gain comes from: POCKET reasons more concisely, so more of its answers finish within the same budget (675 vs 798 of 4,000 samples reached the 16K cap). It uses about 13% fewer tokens per question (geometric mean). The largest gains are on the longest questions, where the parent often runs out of budget before answering.

On MMLU-Pro (2,000 questions) the two builds score the same (87.65% vs 87.95%, within noise), while POCKET's correct answers are shorter: short answers are unchanged, and answers past roughly 700 tokens get 11 to 26% shorter.

These numbers were checked in public. A community member, Dipankar Sarkar, verified the graft byte by byte and reconciled our tables in the discussion here:
https://huggingface.co/posts/SeaWolf-AI/446226611129845

Next: a greedy-decoding comparison to locate where in the reasoning the saving happens (before or after the model first commits to an answer). We will post the results in that thread.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment