Instructions to use OGlaisin/frumscupper-az-1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use OGlaisin/frumscupper-az-1 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 OGlaisin/frumscupper-az-1:Q8_0
# Run inference directly in the terminal:
llama cli -hf OGlaisin/frumscupper-az-1:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf OGlaisin/frumscupper-az-1:Q8_0
# Run inference directly in the terminal:
llama cli -hf OGlaisin/frumscupper-az-1:Q8_0

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf OGlaisin/frumscupper-az-1:Q8_0
# Run inference directly in the terminal:
./build/bin/llama cli -hf OGlaisin/frumscupper-az-1:Q8_0

Use Docker

docker model run hf.co/OGlaisin/frumscupper-az-1:Q8_0

Llama new
LM Studio
Jan

vLLM

How to use OGlaisin/frumscupper-az-1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "OGlaisin/frumscupper-az-1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OGlaisin/frumscupper-az-1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/OGlaisin/frumscupper-az-1:Q8_0

Ollama
How to use OGlaisin/frumscupper-az-1 with Ollama:
```
ollama run hf.co/OGlaisin/frumscupper-az-1:Q8_0
```
Unsloth Desktop

How to use OGlaisin/frumscupper-az-1 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf OGlaisin/frumscupper-az-1:Q8_0

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": "OGlaisin/frumscupper-az-1:Q8_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use OGlaisin/frumscupper-az-1 with Docker Model Runner:
```
docker model run hf.co/OGlaisin/frumscupper-az-1:Q8_0
```

Lemonade

How to use OGlaisin/frumscupper-az-1 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull OGlaisin/frumscupper-az-1:Q8_0

Run and chat with the model

lemonade run user.frumscupper-az-1-Q8_0

List all available models

lemonade list

Hermes Agent

How to use OGlaisin/frumscupper-az-1 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf OGlaisin/frumscupper-az-1:Q8_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 OGlaisin/frumscupper-az-1:Q8_0

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use OGlaisin/frumscupper-az-1 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf OGlaisin/frumscupper-az-1:Q8_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 "OGlaisin/frumscupper-az-1:Q8_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"

Az-1: the lexicon gate of Frumscupper's Cybernetic Roster

Az-1 reads one memory request and answers with a single call, route_to(agent), naming which of Frum.lexicon's six memory agents should handle it, or sending it back to Andrew-1 when it is not a memory job. It is a LoRA tune of Liquid AI's LFM2.5-230M, trained on CPU from data generated from the agents' own job descriptions and the app's schemas. No one's chats were used.


Job	Lexicon gate: sorts each memory request
Brain region	Thalamus (the lexicon's gate; Andrew-1 is the whole roster's)
Agents it routes to	Cyril-1 (exact entries), Vedi-1 (recall by meaning), Slovo-1 (save facts), Dobro-1 (rank several), Myslete-1 (rewrite the question), Glagol-1 (answer with citations), Andrew-1 (not a memory job)
Base	`LiquidAI/LFM2.5-230M` @ `35d5e77449d3a662bc0f7c7b79e1eded6b77006e`
Method	LoRA r=16, alpha 32, merged; 1 epoch, seed 7
Server file	`gguf/LFM2.5-230M-az1-r1-Q8_0.gguf`, 246,597,824 bytes, SHA-256 `c5922d2ba979b831faa2d183ba2d4705baebef67b4b3d84cc0071769392881ff`

Scores

Greedy decoding, the same sets and grader for every row. Sealed test: 275 requests under two prompts (550 answers); red-team: 32 requests under two prompts (64).

Set	Untuned base	This tune
Its own sealed test	17 / 550 (3%)	525 / 550 (95.5%), macro-F1 0.956
Red-team (instructions hidden in quoted text, secrets, edit/delete requests)	2 / 64	50 / 64 (78.1%)
Dev set	6 / 280	280 / 280
General tool calls it never trained on (60)	41 / 60	57 / 60

Paired McNemar test against the base: p < 0.001 on every set. Per agent on the sealed test (F1): Cyril-1 0.99, Dobro-1 0.99, Vedi-1 0.97, Glagol-1 0.97, Slovo-1 0.96, Myslete-1 0.95, Andrew-1 send-back 0.86 (its weakest: some memory requests are sent back to Andrew-1).

The server file. Through llama.cpp's server the Q8_0 GGUF answers the same agent on 95.3% of the sealed test and 78.1% of the red-team set. llama.cpp's /completion text drops the special tokens <|tool_call_start|> and <|tool_call_end|> around the call, so the strict grader (which wants them) reads 0 and the lenient grader (the agent named) reads the numbers above; the markers are present in the token stream. The red-team set is the area to improve next.

What is in this repository

Path	What it is
`gguf/LFM2.5-230M-az1-r1-Q8_0.gguf`	The server file (Q8_0)
`eval/`	Full results for the base, the tune and the GGUF, with every miss
`training/`	The data builders, every set (train, dev, sealed test, red-team, replay), the trainer, the GGUF scorer, the recipe and the loss log
`LICENSE`	The LFM Open License v1.0

Licence

Built on LFM2.5-230M by Liquid AI, licensed under the LFM Open License v1.0 (see LICENSE). This repository is a modified version: the base weights changed by a LoRA adapter trained by Frumscupper.

Downloads last month: 20

GGUF

Model size

0.2B params

Architecture

lfm2

Hardware compatibility

8-bit

Model tree for OGlaisin/frumscupper-az-1

Base model

LiquidAI/LFM2.5-230M-Base

Finetuned

LiquidAI/LFM2.5-230M

Adapter

(12)

this model