qwen3-vl-2b

A Core AI bundle of Qwen/Qwen3-VL-2B-Instruct for the Aether SDK (iOS and macOS 27+).

• Source: Qwen/Qwen3-VL-2B-Instruct at revision 89644892e4d85e24eaac8bacfd4f463576704203, licence Apache-2.0. The licence is included as LICENSE. Qwen/Qwen3-VL-2B-Instruct declares apache-2.0 in its model card but ships no licence file; LICENSE is the canonical text from apache.org.
• Changes from the source: converted from PyTorch to Core AI (.aimodel) by Aether forge (recipe qwen3-vl-2b@3). Weights are int8-linear-perblock32 (8-bit weights). Two assets per variant: the text tower (these weights) and the vision encoder (float16 weights, float32 compute), a separate -vision asset that loads only for image turns. The tokenizer files are the source's own.
• Input: text and images (imageInput); text out.

Variants

Variant	Platform	Arch	Compute	Compiled	Assets	Download
macos-any-gpu	macos	any	gpu	no (specialized on first load)	qwen3_vl_2b-vision.aimodel 814.9 MB, qwen3_vl_2b.aimodel 1.83 GB	2.66 GB
ios-any-gpu	ios	any	gpu	no (specialized on first load)	qwen3_vl_2b-vision.aimodel 814.9 MB, qwen3_vl_2b.aimodel 1.83 GB	2.66 GB

Verification

Every row is a record in verification/ about exactly these bytes (matched by bundle digest). Reference rows are strict T2 passes of the unquantized export on the same fixture, in verification/reference/.

Variant	Tier	Result	Detail	Device	OS build	Compute	Record
ios-any-gpu	T2	pass	18/18 strict; profile quantized-8bit; fixture 7cfe8f0add2a4f6a	iPhone18,2	24A437	target	d6f97d20
ios-any-gpu	T2	pass	7/7 strict; profile quantized-8bit; fixture 26eb0569fc25e644	iPhone18,2	24A437	target	fd089f9d
macos-any-gpu	T2	pass	7/7 strict; profile quantized-8bit; fixture 26eb0569fc25e644	Mac17,6	26A428	gpu	247be5b7
macos-any-gpu	T2	pass	18/18 strict; profile quantized-8bit; fixture 7cfe8f0add2a4f6a	Mac17,6	26A428	gpu	2913acba
macos-any-gpu	T2	pass	6/7 strict; profile quantized-8bit; budgeted: square-colors; fixture 26eb0569fc25e644	Mac17,6	26A428	cpuOnly	d88791d3
macos-any-gpu	T2	pass	18/18 strict; profile quantized-8bit; fixture 7cfe8f0add2a4f6a	Mac17,6	26A428	cpuOnly	f861c82b
macos-any-gpu	T3	pass	vision-shapes-v1; 91.2% vs reference 91.2%; 80 items	Mac17,6	26A428	gpu	72234db1
unquantized reference (not published)	T2	pass	7/7 strict; profile strict; fixture 26eb0569fc25e644	Mac17,6	26A428	gpu	bc7e5f16
unquantized reference (not published)	T2	pass	18/18 strict; profile strict; fixture 7cfe8f0add2a4f6a	Mac17,6	26A428	gpu	d2d7ee33

The quantized profile also requires: T2 strict on the unquantized reference export (met by the reference row).

Use

aether run qwen3-vl-2b --prompt "Hello"
aether run qwen3-vl-2b --image photo.jpg --prompt "What is in this picture?"

import Aether

let aether = try Aether()
let chat = try await aether.chat("qwen3-vl-2b")
let photo = URL(fileURLWithPath: "photo.jpg")
let reply = try await chat.respond(to: ChatMessage(role: .user, content: [
    .image(ImageInput(.url(photo))), .text("What is in this picture?"),
]))
print(reply.text)