qwen3-embedding-0.6b

A Core AI bundle of Qwen/Qwen3-Embedding-0.6B for the Aether SDK (iOS and macOS 27+).

• Source: Qwen/Qwen3-Embedding-0.6B at revision 97b0c614be4d77ee51c0cef4e5f07c00f9eb65b3, licence Apache-2.0. The licence is included as LICENSE. Qwen/Qwen3-Embedding-0.6B 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-embedding-0.6b@2). Weights are int8-linear-perblock32 (8-bit weights). The tokenizer files are the source's own.
• Output: one vector per text (embeddings), for search and similarity; see Embeddings below.

Variants

Variant	Platform	Arch	Compute	Compiled	Assets	Download
macos-any-gpu	macos	any	gpu	no (specialized on first load)	qwen3_embedding_0_6b.aimodel 633.5 MB	649.3 MB
ios-any-gpu	ios	any	gpu	no (specialized on first load)	qwen3_embedding_0_6b.aimodel 633.5 MB	649.3 MB

Embeddings

• Width: 1024 dimensions; Matryoshka: any prefix from 32 to 1024 dimensions (EmbedOptions(dimensions:)), re-normalized.
• Pooling: the last token's hidden state (lastToken); vectors are unit length (normalized in the graph).
• Token limit: 8192 tokens per text, the appended tokens included; longer texts are refused (AE-EMB-001).
• Appended tokens: none beyond the tokenizer's own.
• Queries are rendered as Instruct: {instruction}\nQuery:{text} with the default instruction Given a web search query, retrieve relevant passages that answer the query; documents as {text}.

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	14/14 cases; profile quantized-8bit; cosine ≥ 0.999; fixture d7defd0591ac7299	iPhone18,2	24A446	target	f09227c4
macos-any-gpu	T2	pass	14/14 cases; 14/14 cases; profile quantized-8bit; cosine ≥ 0.999; fixture d7defd0591ac7299	Mac17,6	26A434	target	65fc8aa7
unquantized reference (not published)	T2	pass	14/14 cases; 14/14 cases; profile strict; cosine ≥ 0.999; fixture d7defd0591ac7299	Mac17,6	26A434	target	e1539410

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

Use

aether embed qwen3-embedding-0.6b --query "What is the capital of China?" --text "The capital of China is Beijing."

import Aether

let aether = try Aether()
let embedder = try await aether.embedder("qwen3-embedding-0.6b")
let vectors = try await embedder.embed([.query("What is the capital of China?"), .document("The capital of China is Beijing.")])
print(Embedding.cosine(vectors[0], vectors[1]))