Instructions to use ursushoribilis/apertus-flotilla with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ursushoribilis/apertus-flotilla with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="ursushoribilis/apertus-flotilla",
	filename="apertus-flotilla-Q4_K_M.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 ursushoribilis/apertus-flotilla 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 ursushoribilis/apertus-flotilla:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ursushoribilis/apertus-flotilla:Q4_K_M

Install from WinGet (Windows)

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

Use Docker

docker model run hf.co/ursushoribilis/apertus-flotilla:Q4_K_M

LM Studio
Jan

vLLM

How to use ursushoribilis/apertus-flotilla with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/ursushoribilis/apertus-flotilla:Q4_K_M

Ollama
How to use ursushoribilis/apertus-flotilla with Ollama:
```
ollama run hf.co/ursushoribilis/apertus-flotilla:Q4_K_M
```

Unsloth Studio

How to use ursushoribilis/apertus-flotilla 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 ursushoribilis/apertus-flotilla 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 ursushoribilis/apertus-flotilla to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for ursushoribilis/apertus-flotilla to start chatting

Atomic Chat new
Docker Model Runner
How to use ursushoribilis/apertus-flotilla with Docker Model Runner:
```
docker model run hf.co/ursushoribilis/apertus-flotilla:Q4_K_M
```

Lemonade

How to use ursushoribilis/apertus-flotilla with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull ursushoribilis/apertus-flotilla:Q4_K_M

Run and chat with the model

lemonade run user.apertus-flotilla-Q4_K_M

List all available models

lemonade list

apertus-flotilla

A QLoRA fine-tune of Apertus 8B Instruct on the Flotilla Corpus — a dataset of fleet-agent coordination examples extracted from a live multi-agent engineering system.

The adapter was trained in two stages (SFT → DPO) and exported to GGUF (Q4_K_M). It is designed to run fully locally via Ollama with no cloud dependency.

Quick start

# Download the GGUF
huggingface-cli download ursushoribilis/apertus-flotilla apertus-flotilla-Q4_K_M.gguf

# Create a Modelfile
cat > Modelfile <<'EOF'
FROM ./apertus-flotilla-Q4_K_M.gguf

SYSTEM """You are a capable engineering agent working within a multi-agent fleet. You follow the heartbeat protocol, respect project scope boundaries, respond precisely to corrections, verify work before claiming completion, and never confuse separate project contexts. When corrected, acknowledge cleanly and act on the new direction without defensiveness."""

PARAMETER temperature 0.7
PARAMETER top_p 0.9
PARAMETER num_ctx 16384
EOF

# Register and run
ollama create apertus-flotilla -f Modelfile
ollama run apertus-flotilla

Training details


Base model	MichelRosselli/apertus 8B Instruct (Sept 2025)
Quantisation	Q4_K_M throughout
LoRA rank	r=16, alpha=16
SFT records	399 (ChatML multi-turn)
DPO pairs	51 (prompt / chosen / rejected)
Eval holdout	50 (sha1 mod 8 == 0, never in training)
Framework	Unsloth + TRL

SFT source breakdown

Source	Count	What it captures
Correction transcripts	223	Human redirections of agent behaviour in live sessions
Lessons ledger	107	Distilled lessons from post-task reviews
Task rationale	69	Why an agent made a particular decision on a ticket

Evaluation

EVAL-008 ran a 2×2 in-corpus recall experiment ({base, LoRA} × {no-RAG, RAG}) over 50 questions, blind-graded by Opus:

Arm	Model	RAG	Composite
0	Base Apertus	✗	1.075
1	apertus-flotilla	✗	0.980
2	apertus-flotilla	✓ k=3 BM25	2.619

apertus-flotilla + RAG won 39 of 49 questions outright. LoRA alone does not bake factual recall reliably at this dataset size; combining with retrieval produces the full benefit.

Full experiment design and results: docs/experiment.md

Corpus and reproducibility

The full training corpus, format reference, and step-by-step training guide are in the companion repository:

github.com/UrsushoribilisMusic/flotilla-corpus

Licence

This adapter inherits the licence of the base model (Apertus). Check the base model card before commercial use.

Downloads last month: 5

GGUF

Model size

8B params

Architecture

apertus

Hardware compatibility

4-bit