- Vendor
- Quantization
- MLX 4-bit
- Parameters
- 1B
- Size
- 568.8 MB
$ brew install mirai$ mirai --model mlx-community/gemma-3-1b-it-4bitBenchmarks
Gemma-3
Apple M4 Max 128GB
0.21 s
lower is better ↓
Gemma-3
Apple M4 Max 128GB
199 t/s
higher is better ↑
Gemma-3
Apple M4 Max 128GB
0.75 GB
lower is better ↓
Benchmarked 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
A 4-bit quantized version of Google's Gemma 3 1B Instruct model, converted to the Apple MLX framework for efficient on-device inference on Apple Silicon hardware.
Origin & Architecture
This model is a community conversion of google/gemma-3-1b-it, part of Google's Gemma 3 family of lightweight language models. At 1 billion parameters, it sits at the compact end of the Gemma lineup — designed for fast, resource-friendly text generation while retaining strong instruction-following capabilities. The 4-bit quantization further reduces memory footprint and accelerates inference, making it well-suited for local deployment on Mac laptops and desktops.
Key Details
- Base model: Google Gemma 3 1B Instruct
- Quantization: 4-bit (via `mlx-lm` v0.21.6)
- Framework: Apple MLX
- Task: Text generation (instruction-tuned)
Use Cases
This model is a strong fit for developers and researchers who want a lightweight, responsive chat or instruction-following model running natively on Apple Silicon — no cloud dependency required. Typical applications include:
- Local chatbots and assistants
- Quick prototyping of text generation pipelines
- On-device summarization, rewriting, and Q&A
- Educational exploration of LLM behavior
Considerations
The model is compact by design; for tasks demanding deeper reasoning or broader world knowledge, larger Gemma variants may be more appropriate. Access requires agreeing to Google's Gemma usage license on Hugging Face.