AI · Collection · Hebrish
Sabich v2.3
Sabich v2.3 (2026-10-30): the recommended Hebrish model, every format, its benchmark and demo. hebrish.app/models/sabich
4 models · 2 spaces · 2 datasets
- model Hebrish/Sabich Sabich v2.3 (.nemo, run r13-all). Test set v2: 4.96% code-switching WER, 67.0% Hebrew term recall. Access request.
- model Hebrish/Sabich-sherpa-onnx-int8 v2.3, sherpa-onnx int8: what the Hebrish Android and desktop apps run.
- model Hebrish/Sabich-ONNX v2.3, ONNX fp32 and int8 (onnx-asr layout) for onnx-asr, ONNX Runtime, Voxtype.
- model Hebrish/Sabich-GGUF v2.3, GGUF F16 for transcribe.cpp on a desktop GPU.
- space Hebrish: try Sabich Try v2.3 in your browser.
- dataset Hebrish/English-Hebrew-Codeswitching-ASR-Benchmark Public benchmark. v2.3 is scored on test set v2 (config v2, 2026-10-30).
- space English-Hebrew Code-Switching ASR Leaderboard Leaderboard with v2.3 on test set v2.
- dataset Hebrish/Parakeet-V2-Blend-All The training blend Sabich is trained on.
Copied from Hugging Face on 7 November 2026.