Gemma-3

Name: mlx-community/gemma-3-4b-it-8bit
Author: Google

mlx-community/gemma-3-4b-it-8bit

Vendor: Google
Quantization: MLX 8-bit
Parameters: 4B
Size: 3.9 GB

$ brew install mirai$ mirai --model mlx-community/gemma-3-4b-it-8bit

Benchmarks

0.78 s

lower is better ↓

0.78 s

uzu

0.89 s

mlx

0.73 s

llamacpp

92 t/s

higher is better ↑

92 t/s

uzu

92 t/s

mlx

71 t/s

llamacpp

4.42 GB

lower is better ↓

4.42 GB

18.99 GB

5.21 GB

0.31.2 llama.cpp0.3.23/1,343 input tokens/306 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 Google's Gemma 3 4B Instruct model, converted to the MLX format for efficient inference on Apple Silicon hardware. This community conversion was produced using `mlx-vlm` and is based on the official `google/gemma-3-4b-it` release.

Overview

Gemma 3 4B IT is a compact, instruction-tuned language model from Google's Gemma family. It supports multimodal image-text-to-text tasks, meaning it can accept both images and text as input and generate text responses — making it suitable for visual question answering, image captioning, and general conversational use.

The MLX conversion brings this model natively to Apple's MLX framework, enabling fast on-device inference on Mac computers equipped with M-series chips. The 8-bit quantization reduces memory usage compared to the full-precision model while retaining strong output quality.

Key Details

Base model: Google Gemma 3 4B PT (pretrained), instruction-tuned variant
Architecture: Gemma 3 (transformer-based, multimodal)
Quantization: 8-bit (MLX format)
Pipeline: Image-text-to-text generation
Framework: MLX / mlx-vlm
License: Gemma (Google usage license, gated access required)

Use Cases

Visual understanding: Describe, analyze, or answer questions about images
Conversational AI: Lightweight instruction-following chat on local hardware
Edge deployment: Run a capable multimodal model entirely on-device with Apple Silicon, no cloud dependency needed

Notes

This is a community-maintained conversion. For full model documentation, training details, and evaluation benchmarks, refer to the original Gemma 3 4B IT model card. Access requires agreeing to Google's Gemma usage license via Hugging Face.

Explore all local models