Description
LiquidAI LFM2.5-24B-A2B is a state-of-the-art reasoning model designed to operate efficiently with only 2B active parameters out of a total of 24B. This model is part of the LFM2.5 family and is tailored for fast, private on-device applications. It excels in agentic and tool-use tasks, demonstrating competitive performance against larger models like Qwen3.6-35B-A3B.
Read More
The model is engineered for efficient inference, running up to 1.7 times faster than comparable models on devices such as the Apple M5 Max and NVIDIA B300 GPU. It supports a wide range of languages, including English, Arabic, Chinese, and more, making it versatile for multilingual applications.
LFM2.5-24B-A2B is available in various formats, including FP8 for high-throughput GPU serving, GGUF for CPU inference, and ONNX for cross-platform deployment. This flexibility allows it to be integrated into diverse environments, from cloud to edge to mobile.
The model is particularly suited for applications requiring reasoning and tool-calling capabilities. It supports function calling in a structured four-step process, enhancing its utility in agentic workflows. However, it is not recommended for knowledge-heavy tasks without supplementary retrieval or web search capabilities.
For developers, LFM2.5-24B-A2B offers extensive fine-tuning options, including supervised fine-tuning and direct preference optimization using frameworks like Unsloth and TRL. The model's performance has been benchmarked across various tasks, consistently showing high efficiency and speed.
Overall, LiquidAI LFM2.5-24B-A2B is a robust choice for developers seeking a powerful, efficient reasoning model for on-device applications.
- CategoryAI Models & LLMs
- Pricing ModelOpen source
- DeploymentCloud, On-premises
- PlatformsWeb, Windows, macOS
LiquidAI LFM2.5-24B-A2B Highlights
24B total parameters
2B active parameters
Efficient on-device inference
Supports multiple languages
Tool-calling capabilities
Available in multiple formats
Fine-tuning support
High-speed GPU inference
Getting Started with LiquidAI LFM2.5-24B-A2B
Access model page: Visit the Hugging Face model page
Authenticate: Log in to your Hugging Face account
Configure: Set up the model for your specific use case
Send prompts: Use the model for reasoning tasks
Fine-tune: Customize the model with your data
LiquidAI LFM2.5-24B-A2B's Use Cases
- Multilingual Support
- On-Device Agents
- Tool-Calling Workflows
- Cross-Platform Deployment
- Fine-Tuning Applications
