---
base_model: ermiaazarkhalili/LFM2.5-350M-Function-Calling-xLAM-Unsloth
tags:
- gguf
- llama.cpp
- unsloth
- lfm2.5
- function-calling
- quantized
license: apache-2.0
language:
- en
datasets:
- Salesforce/xlam-function-calling-60k
pipeline_tag: text-generation
---

# LFM2.5-350M-Function-Calling-xLAM-Unsloth — GGUF quantized

GGUF quantizations of [`ermiaazarkhalili/LFM2.5-350M-Function-Calling-xLAM-Unsloth`](https://huggingface.co/ermiaazarkhalili/LFM2.5-350M-Function-Calling-xLAM-Unsloth),
produced via [Unsloth](https://github.com/unslothai/unsloth) + llama.cpp's conversion scripts.

| Field | Value |
|---|---|
| **Source checkpoint** | [`ermiaazarkhalili/LFM2.5-350M-Function-Calling-xLAM-Unsloth`](https://huggingface.co/ermiaazarkhalili/LFM2.5-350M-Function-Calling-xLAM-Unsloth) |
| **Base model** | [`LiquidAI/LFM2.5-350M`](https://huggingface.co/LiquidAI/LFM2.5-350M) |
| **Dataset** | [`Salesforce/xlam-function-calling-60k`](https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k) |
| **Training** | 1 full epoch (effective batch=8) |
| **Final training loss** | 0.6507 (job 36550863, runtime 46m 11s, peak VRAM 5.73 GB) |
| **Conversion** | Unsloth `save_pretrained` → llama.cpp `convert_hf_to_gguf.py` → `llama-quantize` |

## Available quantizations

| File | Bits | Size | Notes |
|---|---|---|---|
| `lfm2.5-350m-function-calling-xlam-unsloth.q2_k.gguf` | 2-bit | 0.16 GB | Smallest; aggressive quality loss |
| `lfm2.5-350m-function-calling-xlam-unsloth.q3_k_m.gguf` | 3-bit | 0.19 GB | Small; noticeable quality loss |
| `lfm2.5-350m-function-calling-xlam-unsloth.q4_k_m.gguf` | 4-bit | 0.23 GB | **Recommended** — best size/quality balance |
| `lfm2.5-350m-function-calling-xlam-unsloth.q5_k_m.gguf` | 5-bit | 0.26 GB | Near-full quality |
| `lfm2.5-350m-function-calling-xlam-unsloth.q6_k.gguf` | 6-bit | 0.29 GB | Very close to Q8_0 at smaller size |
| `lfm2.5-350m-function-calling-xlam-unsloth.q8_0.gguf` | 8-bit | 0.38 GB | Largest; closest to bf16 source |

**Recommended default:** `Q4_K_M`. For maximum fidelity, use `Q8_0`.

## Usage

### llama.cpp

```bash
llama-cli -hf ermiaazarkhalili/LFM2.5-350M-Function-Calling-xLAM-Unsloth-GGUF --jinja -p "Find flights from SFO to NYC on December 25th" -n 256
```

### Ollama

```bash
ollama run hf.co/ermiaazarkhalili/LFM2.5-350M-Function-Calling-xLAM-Unsloth-GGUF:Q4_K_M
```

### llama-cpp-python

```python
from llama_cpp import Llama
llm = Llama.from_pretrained(
    repo_id="ermiaazarkhalili/LFM2.5-350M-Function-Calling-xLAM-Unsloth-GGUF",
    filename="*q4_k_m.gguf",
    n_ctx=2048,
)
out = llm.create_chat_completion(
    messages=[{"role": "user", "content": "Find flights from SFO to NYC on December 25th"}],
    max_tokens=256,
)
print(out["choices"][0]["message"]["content"])
```

## Intended use

For research and non-commercial experimentation. Outputs should be independently verified before any downstream use.

## Limitations

- GGUF quantizations have unavoidable quality loss vs the bf16 source. Use Q5_K_M+ for best fidelity.
- Inherits all limitations of the source merged checkpoint ([`ermiaazarkhalili/LFM2.5-350M-Function-Calling-xLAM-Unsloth`](https://huggingface.co/ermiaazarkhalili/LFM2.5-350M-Function-Calling-xLAM-Unsloth)).

## Citation

```bibtex
@misc{ lfm25_350m_xlam_unsloth_2026_gguf ,
  author = {Ermia Azarkhalili},
  title = { LFM2.5-350M-Function-Calling-xLAM-Unsloth — GGUF quantized },
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/ermiaazarkhalili/LFM2.5-350M-Function-Calling-xLAM-Unsloth-GGUF}}
}
```

---

This lfm2.5 model was trained 2× faster with [Unsloth](https://github.com/unslothai/unsloth) and Hugging Face's TRL library.

[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)