L

LFM2 350M

Developed by LiquidAI
LFM2-350M is a hybrid model developed by Liquid AI, designed for edge AI and device-side deployment, with efficient training and inference capabilities.
Downloads 1,519
Release Time : 7/10/2025

Model Overview

LFM2-350M is a new type of hybrid Liquid model suitable for edge device deployment, supporting fast training and inference, and outperforming models of similar scale in performance.

Model Features

Fast Training and Inference
The training speed is 3 times faster than the previous generation, and the decoding and pre - filling speed on the CPU is 2 times faster than Qwen3.
Excellent Performance
It performs better than models of similar scale in benchmark tests such as knowledge, mathematics, instruction following, and multilingual capabilities.
New Architecture
It adopts a hybrid Liquid model architecture with multiplication gates and short convolutions.
Flexible Deployment
It supports efficient operation on CPU, GPU, and NPU hardware and is suitable for devices such as smartphones, laptops, or vehicles.

Model Capabilities

Text Generation
Multilingual Processing
Instruction Following
Mathematical Reasoning
Tool Invocation

Use Cases

Agent Tasks
Data Extraction
Extract structured data from text.
Creative Writing
Story Generation
Generate short stories or creative texts.
Multi - turn Dialogue
Chat Assistant
A chatbot that supports multi - turn dialogue.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase