Instructions to use tpnlabs/LFM2.5-1.2B-Instruct-TPN-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 tpnlabs/LFM2.5-1.2B-Instruct-TPN-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 tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf tpnlabs/LFM2.5-1.2B-Instruct-TPN-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 tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf tpnlabs/LFM2.5-1.2B-Instruct-TPN-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 tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF:Q4_K_M

Use Docker

docker model run hf.co/tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tpnlabs/LFM2.5-1.2B-Instruct-TPN-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": "tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF:Q4_K_M

Ollama
How to use tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF with Ollama:
```
ollama run hf.co/tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF:Q4_K_M
```
Unsloth Desktop

How to use tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf tpnlabs/LFM2.5-1.2B-Instruct-TPN-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": "tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF with Docker Model Runner:
```
docker model run hf.co/tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF:Q4_K_M
```

Lemonade

How to use tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.LFM2.5-1.2B-Instruct-TPN-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use tpnlabs/LFM2.5-1.2B-Instruct-TPN-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 tpnlabs/LFM2.5-1.2B-Instruct-TPN-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 tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf tpnlabs/LFM2.5-1.2B-Instruct-TPN-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 "tpnlabs/LFM2.5-1.2B-Instruct-TPN-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"

LFM2.5-1.2B-Instruct · TPN-optimised GGUF

Quantised and measured by TPN Workbench from LiquidAI/LFM2.5-1.2B-Instruct @ 0f604ad. Every format is scored against the BF16 reference on next-token KL divergence, perplexity and top-1 agreement.

Full report: https://bench.trueperformancenetwork.com/r/DGiXiKCe9ZoZ9uoncS1v2A

TPN recommendation: Q6_K: Highest-fidelity format that fits 1 GB: 0.963 GB, closeness 0.008.

Files

file	format	size	mean KLD	KL similarity	PPL Δ	top-1
`LFM2.5-1.2B-Instruct-Q4_K_M.gguf`	Q4_K_M	0.73 GB	0.0566	94.5 %	-1.36 %	88.4 %
`LFM2.5-1.2B-Instruct-Q5_K_M.gguf`	Q5_K_M	0.84 GB	0.0185	98.2 %	+0.06 %	93.4 %
`LFM2.5-1.2B-Instruct-Q6_K.gguf`	Q6_K	0.96 GB	0.0069	99.3 %	+0.25 %	95.9 %	TPN recommendation

Lower KLD = closer to the original. K-quants have SIMD kernels in llama.cpp WebAssembly and repacked ARM kernels on phones, so they are the faster pick on those devices.

Measurements

format	size	bpw	mean KLD	PPL Δ	top-1
Q8_0	1.25 GB	8.5	0.0014	+0.61 %	98.0 %
Q6_K	0.96 GB	6.56	0.0069	+0.25 %	95.9 %
Q5_0	0.86 GB	5.5	0.0266	+2.31 %	92.0 %
Q5_0+	0.86 GB	5.5	0.0266	+2.31 %	92.0 %
Q5_K_M	0.84 GB	5.64	0.0185	+0.06 %	93.4 %
Q4_0M+	0.76 GB	4.79	0.0661	+1.00 %	87.7 %
Q4_K_M	0.73 GB	4.77	0.0566	-1.36 %	88.4 %
Q4_0	0.73 GB	4.5	0.0940	+8.18 %	85.3 %
Q4_0+	0.73 GB	4.5	0.0940	+8.18 %	85.3 %
IQ4_XS	0.66 GB	4.25	0.0724	+2.02 %	87.1 %
Q3_K_M	0.60 GB	3.76	0.1582	+5.37 %	80.7 %
Q2_K	0.48 GB	2.86	0.5384	+27.09 %	66.5 %
IQ2_M	0.43 GB	2.62	0.5624	+29.30 %	66.0 %
IQ2_S	0.40 GB	2.38	0.6961	+46.13 %	62.2 %
IQ2_XS	0.40 GB	2.33	0.8351	+75.12 %	58.3 %
IQ2_XXS	0.37 GB	2.09	1.2455	+159.03 %	50.8 %

Task benchmarks

format	ifeval
Q6_K	85.6
Q5_K_M	84.0
Q4_K_M	84.5

Gratis chat eval

32 everyday prompts (creative, instructions, reasoning, knowledge, tasks, de/es/fr, honesty, chat) × 3 samples with the Gratis system prompt and sampling, thinking off, graded 1-10 by Claude Sonnet; overall = mean of category scores ×10. Run-to-run noise ≈ ±2.

file	overall
`LFM2.5-1.2B-Instruct-Q4_K_M.gguf`	61.8
`LFM2.5-1.2B-Instruct-Q5_K_M.gguf`	65.8
`LFM2.5-1.2B-Instruct-Q6_K.gguf`	65.0

Use

llama-server -hf tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF --hf-file LFM2.5-1.2B-Instruct-Q6_K.gguf

Runs fully on-device in Gratis, in the browser or desktop app.

Licence: inherited from the base model (LFM Open License v1.0).

Downloads last month: 515

GGUF

Model size

1B params

Architecture

lfm2

Hardware compatibility

4-bit

5-bit

6-bit

Model tree for tpnlabs/LFM2.5-1.2B-Instruct-TPN-GGUF

Base model

LiquidAI/LFM2.5-1.2B-Base

Finetuned

LiquidAI/LFM2.5-1.2B-Instruct

Quantized

(131)

this model