---

pipeline_tag: text-generation

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

license: other

license_name: lfm1.0

license_link: LICENSE

library_name: llama-cpp

tags:

• liquid
• lfm2.5
• edge
• moe
• gguf
• nvfp4

language:

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

---

LFM2.5-24B-A2B NVFP4 - GGUF

NVFP4 GGUF conversions derived from the LFM2.5-24B-A2B model, providing highly efficient 4-bit floating-point quantization for reasoning and agentic tasks.

This repository provides NVFP4 (NVIDIA 4-bit floating-point) GGUF files, making the model usable with llama.cpp and other GGUF-compatible inference engines.

Model Overview

LFM2.5-24B-A2B is a 24B-parameter reasoning model with only 2B active parameters, built for fast, private agents on-device. It is the largest and final model in the LFM2.5 family, delivering best-in-class agentic and tool-use performance with MoE speed.

• Best-in-class agent: Competitive with Qwen3.6-35B-A3B, a model with ~50% more total and active parameters, on agentic and tool-use tasks, scoring 48.09 on LiveClaw versus 38.19.
• Reasoning at MoE speed: Always thinks before it answers. With only 2B active parameters per token, each reasoning token is cheap.
• Efficient inference: Runs up to 1.7× faster than comparable MoEs on-device.

Find more information about LFM2.5-24B-A2B in our blog post.

Model Architecture

| Property | Value |

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

| Architecture | LFM2.5 (Mixture of Experts) |

| Total Parameters | 24.0B |

| Active Parameters | ~2.45B |

| Number of Layers | 40 (30 double-gated short convolution blocks + 10 GQA) |

| Experts | 64 per MoE block, top-4 routing |

| Training Budget | 36 trillion tokens |

| Vocabulary Size | 128,000 |

| Context Length | 131,072 tokens |

| Languages | English, Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, Spanish, Vietnamese, Thai, Indonesian, Hindi, Russian, Polish |

GGUF Files

| File | Format | Description |

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

| lfm2.5-24b-a2b-nvfp4.gguf | NVFP4 | NVFP4 quantized model - optimized for efficiency |

> About NVFP4: NVFP4 is NVIDIA's 4-bit floating-point format (E2M1). It offers significant memory savings while preserving quality close to bfloat16, and is accelerated on NVIDIA Blackwell GPUs via llama.cpp.

Usage

llama.cpp (CLI)

# Text generation / Reasoning inference
./llama-cli \
  -m lfm2.5-24b-a2b-nvfp4.gguf \
  -p "Solve the following problem step by step..." \
  --temp 0.1 --top-k 50 --repeat-penalty 1.05

llama-server (OpenAI-compatible API)

# API server
./llama-server \
  -m lfm2.5-24b-a2b-nvfp4.gguf \
  --host [IP_ADDRESS] --port 8080

Generation Parameters

Recommended parameters for LFM2.5-24B-A2B:

| Parameter | Value |

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

| Temperature | 0.1 |

| Top-K | 50 |

| Repetition Penalty | 1.05 |

Key Features

• Reasoning at MoE Speed: Thinks before answering with low active parameter count per token.
• Agent-Optimized: Exceptional performance on tool-use and agentic tasks.
• On-Device Ready: Optimized for high throughput and memory efficiency.

License

lfm1.0