LFM 2
LiquidAI/LFM2-700M- Vendor
- LiquidAI
- Parameters
- 700M
- Size
- 1.4 GB
$ brew install mirai$ mirai --model LiquidAI/LFM2-700MBenchmarks
LFM 2
Apple M4 Max 128GB
0.14 s
lower is better ↓
uzu0.5.14vs
0.31.2vs
llama.cpp0.3.23LFM 2
Apple M4 Max 128GB
290 t/s
higher is better ↑
uzu0.5.14vs0.31.2vsllama.cpp0.3.23LFM 2
Apple M4 Max 128GB
1.52 GB
lower is better ↓
uzu0.5.14vs0.31.2vsllama.cpp0.3.23Benchmarked 7 Aug 2026
Integrate 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
LFM2-700M
LFM2-700M is a 742M-parameter hybrid language model from Liquid AI, purpose-built for edge AI and on-device deployment. Part of the second-generation LFM2 family, it combines a novel architecture of multiplicative gates and short convolutions — specifically 10 double-gated short-range LIV convolution blocks and 6 grouped query attention (GQA) blocks — to deliver strong quality at minimal footprint.
Key Strengths
- Optimized for the edge — Runs efficiently on CPU, GPU, and NPU hardware, making it suitable for smartphones, laptops, and vehicles. LFM2 achieves roughly 2× faster decode and prefill on CPU compared to Qwen3.
- Competitive quality — Outperforms similarly-sized models across knowledge, math, instruction following, and multilingual benchmarks.
- Multilingual — Supports English, Arabic, Chinese, French, German, Japanese, Korean, and Spanish.
- Tool use built in — Natively supports structured function calling via special tokens and JSON schema definitions.
The model supports a 32,768-token context length, uses bfloat16 precision, and was trained on 10 trillion tokens (~75% English, 20% multilingual, 5% code). Training leveraged knowledge distillation from LFM1-7B, large-scale supervised fine-tuning, custom DPO, and iterative model merging.
Performance
LFM2-700M delivers strong results relative to its size class, scoring 49.9 on MMLU, 72.23 on IFEval, and 46.4 on GSM8K — competitive with or exceeding models like Llama-3.2-1B-Instruct and Qwen3-0.6B.
CPU inference throughput is a particular highlight, with LFM2 models outpacing alternatives in both ExecuTorch and llama.cpp runtimes.
Recommended Use
Liquid AI recommends fine-tuning LFM2-700M on narrow use cases for best results. It is well suited for agentic tasks, data extraction, RAG, creative writing, and multi-turn conversations. GGUF quantized checkpoints are also available for llama.cpp deployment. Compatible with Hugging Face Transformers v4.55+ and vLLM v0.10.2+.