0.0.15

Faster similarity math with Accelerate 🙌

What's Changed

• Added AMX acceleration by @johnbean393 in #39

New Contributors

• @johnbean393 made their first contribution in #39

Full Changelog: 0.0.14...0.0.15

0.0.14

What's Changed

• Fix load index by @buhe in #36
  • loadIndex no longer uses addItems and simply assigns the indexItems directly, which should should speed loading 🏎️ 💨

New Contributors

• @Kilo-Loco made their first contribution in #33

Full Changelog: 0.0.13...0.0.14

0.0.13

What's Changed

• Include token_type_ids when building model inputs by @weswalla in #32

New Contributors

• @weswalla made their first contribution in #32

Full Changelog: 0.0.12...0.0.13

0.0.12

What's Changed

• Fix typo by @buhe in #29
• fix: Add completion callback to addItems(_:items) by @brn in #31

New Contributors

• @buhe made their first contribution in #29
• @brn made their first contribution in #31

Full Changelog: 0.0.11...0.0.12

0.0.11

What's Changed

• Additional options for native embeddings by @BernhardEisvogel in #23
• Top k improved search speed by @BernhardEisvogel in #25

New Contributors

• Huge thanks to @BernhardEisvogel for his contributions in this release!

Full Changelog: 0.0.10...0.0.11

0.0.10

Add multilingual native embeddings class NativeContextualEmbeddings

• This new embeddings class uses the latest native embeddings from Apple called NLContextualEmbeddings #22. Prompted by discussion in issue #21 with @aehlke.
• Usage:

        let similarityIndex = await SimilarityIndex(
            model: NativeContextualEmbeddings()
            metric: CosineSimilarity()
        )

Full Changelog: 0.0.9...0.0.10

0.0.9

Remove fatalerror from setupDimension

• Addresses issue #19 where the app was crashing if setupDimension failed.
• Includes other various cleanup

Full Changelog: 0.0.8...0.0.9

0.0.8

Adds BinaryStore

You can now save your index in binary format

• BinaryStore is 4.3x more efficient in disk-storage space than JsonStore. It compresses the data as well.
• Removed some fatal errors
• SimilarityIndex.saveIndex and .loadIndex are now public

Contributors

• @michaeljelly with PR #16

Full Changelog: 0.0.7...0.0.8

0.0.7

Adds RecursiveCharacterSplitter

Contributed by: @LexiestLeszek
PR: #14

Found the description on LangChain website and created my implementation, seems to be working. It will split documents recursively by different characters - starting with "\n\n", then "\n", then " ". This is nice because it will try to keep all the semantically relevant content in the same place for as long as possible.

Orig from langchain: https://js.langchain.com/docs/modules/indexes/text_splitters/examples/recursive_character

0.0.6

Lowered min os versions for overall package

• Distilbert model still requires macOS 13 and iOS 16 due to quantization.