v0.7.0

Announcing React Native ExecuTorch v0.7.0 🌟

If you’re tired of robotic-sounding speech synthesizers, this one’s for you!

What’s new?

• Text to speech: Support for generating speech using Kokoro docs
• Reduced bundle size: We switched a library for tokenization, resulting in 64% reduction in package size
• OCR file sizes improvement: Previously, for the OCR pipeline to work, you needed 4 different model files. We made changes to use just one file, reducing the total file sizes by 3x (313MB -> 101MB) docs
• Improved error handling: The errors thrown from our library are now assigned a code property, allowing for fine-grained error handling docs
• LLMs: new stats API (getPromptTokenCount, getTotalTokenCount)
• LLMs: generate() API now returns a string, instead of void docs
• Our documentation now includes an API reference section
• General bug-fixes

Breaking changes ⚠️

• Removed responseCallback() in LLM APIs
• OCR now takes a single modelSource, instead of 3.
• Removed constants for VERTICAL_OCR. From now, only OCR_* constants should be used for both OCR and vertical OCR.
• Passing waveforms as number[] in Speech to Text is now not supported anymore. Please use Float32Array instead.
• TokenizerModule will now throw RnExecutorchError when idToToken or tokenToIdFails (for example when you pass a token outside of the vocab)
• The error property in all of the hooks now is of type RnExecutorchError instead of string.

v0.6.0

Announcing React Native ExecuTorch v0.6.0 🫶🏻

We are excited to release v0.6.0! This version brings some major performance improvements and introduces powerful new generative capabilities including Stable Diffusion, VAD and improved Whisper support.

What’s new?

• Major performance improvements: see benchmarks for more details
• Text to Image: Support for Stable Diffusion is here! docs
• Voice Activity Detection: Detect speech segments with FSMN-based VAD implementation. docs
• Quantized Whisper: We now support quantized Whisper models, resulting in 4x smaller file sizes without significant accuracy tradeoffs. docs
• LLM Control: You can now configure your generation with configurable temperature and topp parameters. docs
• ExecuTorch Runtime: Bumped to v1.0.0 to match upstream stability and features.
• Android Support: Added support for 16kB memory page sizes.
• General bug fixes

Breaking Changes ⚠️

Expo SDK Upgrade Required
React Native ExecuTorch now depends on the latest expo-file-system API changes. You must upgrade your project to Expo SDK 54 to use this version.

0.5.15

What's Changed

Fixed bug on android where number of threads in threadpools could be set to 0 in effect freezing the library. (issue should only impact simulators)

Full Changelog: v0.5.14...v0.5.15

0.5.13

What's Changed

• Remove CocoaPods cache step from iOS build workflow by @msluszniak in #654
• Fixed bug impacting quality of llms responses and sometimes causing models to keep generating until max number of tokens is reached.

Full Changelog: v0.5.12...v0.5.13

v0.4.10

What's Changed

• Fixed bug impacting quality of llms responses and sometimes causing models to keep generating until max number of tokens is reached.

Full Changelog: v0.4.9...v0.4.10

0.5.12

Fixed bug where special characters resulted in malformed utf-8 encodings when using both usSpeechToText and SpeechToTextModule API.

0.5.11

What's Changed

• chore: 0.5.11 by @mkopcins in #649

Full Changelog: v0.5.10...v0.5.11

0.4.9

Changed jitpack target.

v0.5.10

What's Changed

• chore: readme update by @p3t3rzb in #641

Full Changelog: v0.5.9...v0.5.10

v0.5.9

What's Changed

• fix: useExecutorchModule not working by @jakmro in #637

Full Changelog: v0.5.8...v0.5.9