LFM 2
LiquidAI/LFM2-350M- Vendor
- LiquidAI
- Parameters
- 350M
- Size
- 680.8 MB
$ brew install mirai$ mirai --model LiquidAI/LFM2-350MIntegrate with SDK
https://github.com/trymirai/uzu-swift
| 1 | import Foundation |
| 2 | import Uzu |
| 3 | |
| 4 | public func runChat() async throws { |
| 5 | let engineConfig = EngineConfig.create() |
| 6 | let engine = try await Engine.create(config: engineConfig) |
| 7 | |
| 8 | guard let model = try await engine.model(identifier: "alibaba:qwen3.5:0.8b:mirai:mirai-m:4") else { |
| 9 | return |
| 10 | } |
| 11 | for try await update in try await engine.download(model: model).iterator() { |
| 12 | print(String(format: "\r\u{001B}[2KDownload progress: %.2f%%", update.progress() * 100), terminator: "") |
| 13 | fflush(stdout) |
| 14 | } |
| 15 | print() |
| 16 | |
| 17 | let messages = [ |
| 18 | ChatMessage.system().withText(text: "You are a helpful assistant"), |
| 19 | ChatMessage.user().withText(text: "Tell me a short, funny story about a robot") |
| 20 | ] |
| 21 | let session = try await engine.chat(model: model, config: .create()) |
| 22 | let stream = await session.replyWithStream(input: messages, config: .create()) |
| 23 | var message: ChatMessage? = nil |
| 24 | for try await update in stream.iterator() { |
| 25 | switch update { |
| 26 | case .replies(let replies): |
| 27 | let reply = replies.last |
| 28 | message = reply?.message |
| 29 | print("Generated tokens: \(reply?.stats.tokensCountOutput ?? 0)") |
| 30 | case .error(let error): |
| 31 | print("Error: \(error)") |
| 32 | } |
| 33 | } |
| 34 | print("Reasoning: \(message?.reasoning() ?? "empty")") |
| 35 | print("Text: \(message?.text() ?? "empty")") |
| 36 | } |
| 37 | |
Details
LFM2-350M
LFM2-350M is the smallest model in Liquid AI's second-generation LFM2 family, a series of hybrid language models purpose-built for edge AI and on-device deployment. With roughly 354 million parameters, it delivers a compelling balance of quality, speed, and memory efficiency for resource-constrained environments.
Architecture & Training
LFM2-350M introduces a novel hybrid architecture combining multiplicative gates and short convolutions — specifically 10 double-gated short-range LIV convolution blocks and 6 grouped query attention (GQA) blocks across 16 layers. It supports a 32,768-token context window with a 65,536-token vocabulary in bfloat16 precision.
The model was trained on 10 trillion tokens (approximately 75% English, 20% multilingual, 5% code) using knowledge distillation from LFM1-7B, large-scale supervised fine-tuning, custom DPO with length normalization, and iterative model merging. It supports eight languages: English, Arabic, Chinese, French, German, Japanese, Korean, and Spanish.
Performance
LFM2 models outperform similarly sized competitors across knowledge, math, instruction following, and multilingual benchmarks. The family also achieves up to 2× faster decode and prefill on CPU compared to Qwen3.
Inference throughput comparisons demonstrate strong performance across CPU runtimes:
Best Use Cases
Due to its compact size, Liquid AI recommends fine-tuning LFM2-350M on narrow tasks for best results. It is well suited for agentic workflows, data extraction, RAG pipelines, creative writing, and multi-turn conversations. It also supports structured tool use via JSON function definitions. Deployment is flexible across CPU, GPU, and NPU hardware — ideal for smartphones, laptops, and vehicles.
Compatible with Hugging Face Transformers (v4.55+), vLLM, and llama.cpp via GGUF. Licensed under LFM Open License v1.0.