---
license: other
license_name: lfm1.0
license_link: https://huggingface.co/LiquidAI/LFM2.5-230M/blob/main/LICENSE
base_model: hauser458original/lfm2.5-230m-code-math
tags:
  - lfm2
  - lfm2.5
  - liquid
  - code
  - math
  - gguf
  - llama.cpp
language:
  - en
pipeline_tag: text-generation
---

# LFM2.5-230M-Code-Math-GGUF

GGUF quantized versions of [hauser458original/lfm2.5-230m-code-math](https://huggingface.co/hauser458original/lfm2.5-230m-code-math), a code/math-focused fine-tune of [LiquidAI/LFM2.5-230M](https://huggingface.co/LiquidAI/LFM2.5-230M) (instruct). See the base fine-tune's model card for full training details, evaluation notes, and known limitations.

For use with [llama.cpp](https://github.com/ggml-org/llama.cpp), [Ollama](https://ollama.com/), [LM Studio](https://lmstudio.ai/), or any other GGUF-compatible runtime.

## Files

| File | Quantization | Approx. size | Notes |
|---|---|---|---|
| `lfm2.5-230m-code-math-F16.gguf` | F16 | ~460 MB | Full precision, largest, highest fidelity |
| `lfm2.5-230m-code-math-Q8_0.gguf` | Q8_0 | ~245 MB | Near-lossless, good default if size isn't a concern |
| `lfm2.5-230m-code-math-Q5_K_M.gguf` | Q5_K_M | ~165 MB | Good balance of size/quality |
| `lfm2.5-230m-code-math-Q5_K_S.gguf` | Q5_K_S | ~155 MB | Slightly smaller than Q5_K_M, marginal quality trade-off |
| `lfm2.5-230m-code-math-Q4_K_M.gguf` | Q4_K_M | ~135 MB | Smallest here, most aggressive quantization, best for constrained/edge devices |

(Sizes are approximate — check actual file sizes in the repo.)

## Usage

### llama.cpp
```bash
./llama-cli -m lfm2.5-230m-code-math-Q5_K_M.gguf -p "Write a Python function to check if a number is prime."
```

### Ollama
```bash
ollama run hf.co/hauser458original/lfm2.5-230m-code-math-GGUF:Q5_K_M
```

### LM Studio
Search for `hauser458original/lfm2.5-230m-code-math-GGUF` in the LM Studio model browser, or download a `.gguf` file directly and load it manually.

## Which quant should I use?

- **Q4_K_M**: smallest footprint, best for very constrained devices (older phones, low-RAM edge hardware). Some quality loss vs. higher quants.
- **Q5_K_S / Q5_K_M**: good middle ground — recommended default for most laptop/desktop CPU inference.
- **Q8_0**: near-lossless, use if you have the RAM/storage headroom and want output as close as possible to the original safetensors model.
- **F16**: full precision GGUF, only needed if you plan to re-quantize yourself or want the highest possible fidelity in llama.cpp.

## License

Inherits the [LFM Open License v1.0](https://huggingface.co/LiquidAI/LFM2.5-230M/blob/main/LICENSE) from the base model.