---

license: other

license_name: lfm1.0

license_link: LICENSE

base_model: LiquidAI/LFM2.5-24B-A2B

base_model_relation: quantized

library_name: llama.cpp

pipeline_tag: text-generation

language:

• en
• ar
• zh
• fr
• de
• ja
• ko
• es
• pt
• it
• vi
• th
• id
• hi
• ru
• pl

tags:

• gguf
• llama.cpp
• lfm2.5
• quantized
• text-generation

---

LFM2.5-24B-A2B GGUF

Community GGUF quantizations of LiquidAI/LFM2.5-24B-A2B.

<div align="center" style="background-color:#f59e0b;color:#ffffff;padding:16px 20px;border-radius:10px;line-height:1.7;">

☕ If this GGUF made your day easier, a coffee would make mine.<br>

<a href="https://ko-fi.com/ngquocvinh" style="color:#ffffff;"><strong style="color:#ffffff;">Send a coffee ☕</strong></a><br>

I build and test these releases myself. Your coffee helps keep me going.<br>

Thank you for supporting this work.

</div>

About LFM2.5-24B-A2B

LFM2.5-24B-A2B is Liquid AI's text-only reasoning Mixture-of-Experts model. It has 24 billion total parameters and 2.45 billion active parameters, with 40 layers, 64 experts per MoE block, top-4 routing, and a 131,072-token context window. The upstream card lists English, Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, Spanish, Vietnamese, Thai, Indonesian, Hindi, Russian, and Polish. It describes the source model as a reasoning model with tool use. This release's runtime checks cover short text generation; long-context and tool-call behavior have not been evaluated here.

This release targets llama.cpp and compatible GGUF runtimes. See the official model card for the model's capabilities and usage guidance.


Benchmark figure from the upstream model card; results shown are upstream measurements.

Quick start

After downloading LFM2.5-24B-A2B-Q4_K_M.gguf and the included chat_template.jinja, run:

llama-cli \
  -m LFM2.5-24B-A2B-Q4_K_M.gguf \
  --chat-template-file chat_template.jinja \
  --jinja \
  -p 'Answer briefly in English: What is GGUF and why is it useful for running language models locally?' \
  -n 128 -c 4096 -ngl 99 \
  --temp 0.1 --top-k 50 --repeat-penalty 1.05

Choose a context size that fits your available memory. The 131,072-token context is the model's supported maximum; using it requires substantially more memory than this example.

Reproducibility and validation

Each GGUF is built directly from the pinned official BF16 GGUF source with a model-specific imatrix. Files are added to this repository after llama.cpp loads them and produces a non-empty response. Exact file sizes in decimal GB, including the BF16 reference and imatrix, and per-file validation status are recorded in reproducibility/artifact-measurements.tsv; SHA-256 checksums are in SHA256SUMS.txt, and the build inputs and commands are summarized in reproducibility/manifest.md.

Size measurements

Percentages compare each file's size with the BF16 source (47.962095 GB). Relative size (%) = file size / BF16 size × 100. This percentage measures storage size, not model quality or fidelity.

| Quantization | Size (GB) | Relative size vs BF16 (%) |

| --- | ---: | ---: |

| BF16 source | 47.962095 | 100.00% |

| Q8_0 | 25.493733 | 53.15% |

| Q6_K | 19.689406 | 41.05% |

| Q5_K_M | 17.029603 | 35.51% |

| Q5_K_S | 16.541204 | 34.49% |

| Q4_K_M | 14.526259 | 30.29% |

| Q4_K_S | 13.660242 | 28.48% |

| IQ4_XS | 12.837437 | 26.77% |

| Q3_K_L | 12.428312 | 25.91% |

| Q3_K_M | 11.465719 | 23.91% |

| Q3_K_S | 10.429988 | 21.75% |

| IQ3_XXS | 9.282732 | 19.35% |

| Q2_K | 8.809758 | 18.37% |

| IQ2_XS | 7.090126 | 14.78% |

| IQ2_XXS | 6.370082 | 13.28% |

| TQ2_0 | 6.355845 | 13.25% |

| IQ1_M | 5.471666 | 11.41% |

| Q1_0 | 3.578019 | 7.46% |

No fixed hold-out fidelity comparison is published in this release. Size and runtime compatibility checks do not measure next-token fidelity.

License and attribution

The upstream model is released under the LFM Open License v1.0. Review its terms, including the commercial-use conditions, before using the model. These are community GGUF quantizations, not an official Liquid AI release or endorsement.