Instructions to use ray0rf1re/lfm2.5-ADA-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries
PEFT
How to use ray0rf1re/lfm2.5-ADA-GGUF with PEFT:
```
Task type is invalid.
```
Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use ray0rf1re/lfm2.5-ADA-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 ray0rf1re/lfm2.5-ADA-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ray0rf1re/lfm2.5-ADA-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 ray0rf1re/lfm2.5-ADA-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ray0rf1re/lfm2.5-ADA-GGUF:Q4_K_M

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 ray0rf1re/lfm2.5-ADA-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf ray0rf1re/lfm2.5-ADA-GGUF:Q4_K_M

Use Docker

docker model run hf.co/ray0rf1re/lfm2.5-ADA-GGUF:Q4_K_M

Llama new
LM Studio
Jan
Ollama
How to use ray0rf1re/lfm2.5-ADA-GGUF with Ollama:
```
ollama run hf.co/ray0rf1re/lfm2.5-ADA-GGUF:Q4_K_M
```
Unsloth Desktop

How to use ray0rf1re/lfm2.5-ADA-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ray0rf1re/lfm2.5-ADA-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": "ray0rf1re/lfm2.5-ADA-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use ray0rf1re/lfm2.5-ADA-GGUF with Docker Model Runner:
```
docker model run hf.co/ray0rf1re/lfm2.5-ADA-GGUF:Q4_K_M
```

Lemonade

How to use ray0rf1re/lfm2.5-ADA-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull ray0rf1re/lfm2.5-ADA-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.lfm2.5-ADA-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use ray0rf1re/lfm2.5-ADA-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 ray0rf1re/lfm2.5-ADA-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 ray0rf1re/lfm2.5-ADA-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use ray0rf1re/lfm2.5-ADA-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ray0rf1re/lfm2.5-ADA-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 "ray0rf1re/lfm2.5-ADA-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"

lfm2.5-ADA-GGUF

GGUF quantizations of ray0rf1re/lfm2.5-ADA-merged — the ray0rf1re/lfm2.5-ADA LoRA adapter merged into its base model LiquidAI/LFM2.5-1.2B-Thinking — converted and quantized with llama.cpp.

Files

Quant	Size	Notes
Q4_K_M	697.04 MB	Smallest / fastest. Best choice for low-RAM or CPU-only inference (e.g. this pipeline's target, an i7-7660U). Some quality loss vs f16, usually not noticeable in casual chat.
Q5_K_M	804.29 MB	Middle ground — noticeably better quality than Q4_K_M for a modest size/speed cost. Good default if you have the RAM to spare.
Q6_K	918.24 MB	Largest / highest quality of the three, very close to f16 output. Use this if quality matters more than footprint or speed.

If you're not sure which to grab: Q4_K_M for speed/low RAM, Q6_K if you want quality closest to the unquantized model, Q5_K_M as a middle ground.

Usage (llama.cpp)

# CLI chat
llama-cli -m lfm2.5-ADA-GGUF-Q4_K_M.gguf -p "Hello!" -cnv

# OpenAI-compatible local server
llama-server -m lfm2.5-ADA-GGUF-Q4_K_M.gguf -c 4096

Also works directly in LM Studio, Ollama (via a Modelfile pointing at the .gguf), and any other llama.cpp-based runtime.

Provenance

Base model: LiquidAI/LFM2.5-1.2B-Thinking
LoRA adapter: ray0rf1re/lfm2.5-ADA
Merged full-precision model: ray0rf1re/lfm2.5-ADA-merged
Quantized with: llama.cpp, convert_hf_to_gguf.py (f16) → llama-quantize

Persona benchmark

Automated persona-fidelity benchmark against same-size reference models (see persona_bench.py for methodology). Heuristic score is free/deterministic keyword+repetition scoring; no LLM-judge pass was run for this benchmark.

Model	Avg heuristic (0-1)	Degenerate outputs
lfm2.5-ADA-GGUF-Q4_K_M (ours)	0.558	0 / 8
lfm2.5-ADA-GGUF-Q5_K_M (ours)	0.532	0 / 8
lfm2.5-ADA-GGUF-Q6_K (ours)	0.557	0 / 8
LFM2.5-1.2B-Thinking (base)	0.731	0 / 8
LFM2.5-350M	0.618	0 / 8
Qwen3-0.6B (~0.8B params)	0.736	0 / 8