LFM 2

Name: mlx-community/LFM2-350M-8bit
Author: LiquidAI

mlx-community/LFM2-350M-8bit

Vendor: LiquidAI
Quantization: MLX 8-bit
Parameters: 350M
Size: 364.0 MB

$ brew install mirai$ mirai --model mlx-community/LFM2-350M-8bit

Benchmarks

0.07 s

lower is better ↓

0.07 s

uzu

0.09 s

mlx

0.09 s

llamacpp

755 t/s

higher is better ↑

755 t/s

uzu

689 t/s

mlx

486 t/s

llamacpp

0.48 GB

lower is better ↓

0.48 GB

18.99 GB

0.46 GB

0.31.2 llama.cpp0.3.23/1,339 input tokens/512 output tokens

Benchmarked 7 Aug 2026

Integrate with SDK

Choose framework

Run the following command to install Mirai SDK

https://github.com/trymirai/uzu-swift

Apply code

1import Foundation2import Uzu34public 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        return10    }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? = nil24    for try await update in stream.iterator() {25        switch update {26        case .replies(let replies):27            let reply = replies.last28            message = reply?.message29            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

1Choose framework

2Run the following command to install Mirai SDK

https://github.com/trymirai/uzu-swift

3Apply code

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

An 8-bit quantized version of LiquidAI's LFM2-350M, converted to Apple's MLX format for efficient on-device inference on Apple Silicon hardware. This community conversion makes Liquid AI's compact language model readily accessible for local text generation workflows on Mac.

Origin & Architecture

LFM2-350M is part of Liquid AI's second-generation Liquid Foundation Model family, designed specifically for edge deployment. At just 350 million parameters, it targets scenarios where low latency, small memory footprint, and on-device privacy are priorities. The 8-bit quantization further reduces the model's memory requirements while preserving practical output quality — ideal for resource-constrained environments.

Multilingual Support

Despite its compact size, LFM2-350M supports eight languages: English, Arabic, Chinese, French, German, Japanese, Korean, and Spanish, making it a versatile choice for multilingual text generation tasks at the edge.

Key Details

Base model: LiquidAI/LFM2-350M
Quantization: 8-bit
Framework: MLX (via `mlx-lm` v0.26.0)
Task: Text generation
License: LFM 1.0 (custom)

Best For

Local text generation on Apple Silicon Macs
Edge and on-device language model experimentation
Lightweight multilingual generation where full-scale models are impractical
Developers exploring Liquid AI's novel architecture in a Mac-native runtime

This conversion is maintained by the mlx-community and provides a straightforward path to running one of the smallest capable multilingual models directly on Apple hardware without cloud dependencies.

Explore all local models