---
license: other
license_name: lfm1.0
license_link: https://huggingface.co/LiquidAI/LFM2.5-350M/blob/main/LICENSE
pipeline_tag: text-generation
tags:
- liquid
- lfm2.5
- gguf
- ollama
- edge
- conversational
base_model: LiquidAI/LFM2.5-350M
---

# LiquidAI LFM2.5-350M (Instruct) - GGUF (Q4_K_M)

This repository provides the quantized **Q4_K_M GGUF** weights for **[LiquidAI/LFM2.5-350M](https://huggingface.co/LiquidAI/LFM2.5-350M)**, configured for direct 1-click execution in **Ollama**, **llama.cpp**, and local edge devices.

LFM2.5-350M is a hybrid architecture developed by Liquid AI combining double-gated short convolutions with structured attention for near-linear computational scaling and low memory footprint.

---

## ⚡ Direct Ollama Run (1-Line Command)

You can run this model directly via Ollama without manually downloading any files:

```bash
ollama run hf.co/jamesatron1512/LFM2.5-350M-GGUF
```

Or specify the quantization tag explicitly:

```bash
ollama run hf.co/jamesatron1512/LFM2.5-350M-GGUF:Q4_K_M
```

---

## 🚀 Model Details

- **Parameters**: 350 Million
- **Precision**: Q4_K_M (Quantized 4-bit)
- **File Size**: ~219 MB
- **Context Length**: Up to 128k tokens (default 4096 in Modelfile)
- **Chat Template**: ChatML format (`<|im_start|>user ... <|im_end|>`)
- **System Prompt**: Supported via template, default is left clean to prevent fixation on small parameter counts.

---

## 💻 Python API Usage via Ollama

```python
import requests

response = requests.post(
    "http://localhost:11434/api/generate",
    json={
        "model": "hf.co/jamesatron1512/LFM2.5-350M-GGUF",
        "prompt": "Explain quantum computing in two sentences.",
        "stream": False,
        "options": {
            "temperature": 0.7,
            "top_p": 0.9,
            "num_predict": 128
        }
    }
)

print(response.json()["response"])
```