Models · NVIDIA · Parakeet TDT 0.6B v2
Jachnun main
Parakeet fine-tune: English with Hebrew words. The main Hebrish model and what the apps use. Code-switching fine-tune, every on-device format.
At a glance
| Base model | nvidia/parakeet-tdt-0.6b-v2 (NVIDIA) |
| Current version | Jachnun v1.0, released 2026-10-19 |
| Licence | CC-BY-4.0 · use governed by the terms of use |
| Used by | The Hebrish apps, the browser demo |
| DOIs | 10.57967/hf/11105 10.57967/hf/11104 10.57967/hf/10995 |
| Full model card | results, recipe, limitations |
Formats and quants
| Repository | Format and files | Version | For | DOI |
|---|---|---|---|---|
Hebrish/Jachnun main | .nemo (NeMo) .nemo 2.47 GB, plus evaluations, figures and training configs · access request | v1.0 | GPU inference and further fine-tuning in NVIDIA NeMo | none yet |
Hebrish/Jachnun-sherpa-onnx-int8 | sherpa-onnx int8 encoder, decoder, joiner, about 661 MB · access request | v1.0 | Android, Linux, Windows; what the Hebrish apps run | none yet |
Hebrish/Jachnun-ONNX | ONNX fp32 and int8 int8 encoder 652 MB; fp32 encoder 2.44 GB · access request | v1.0 | onnx-asr, ONNX Runtime, the browser demo (parakeet.js) | 10.57967/hf/11105 |
Hebrish/Jachnun-GGUF | GGUF Q8_0 730 MB, Q4_K_M 475 MB, F16 1.24 GB · access request | v1.0 | transcribe.cpp (Offline Voice Input, Handy) | none yet |
Hebrish/Jachnun-MLX | MLX safetensors 2.47 GB · access request | PoC-3 | parakeet-mlx on Apple silicon | 10.57967/hf/11104 |
danielrosehill/Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching PoC record | .nemo, GGUF, ONNX, sherpa-onnx Every proof-of-concept checkpoint and v1.0, each at a pinned revision · access request | PoC-1 to v1.0 | Where the model was first published; the proof-of-concept record | 10.57967/hf/10995 |
Hugging Face issues a DOI per repository, so each format can be cited on its own. Repositories, sizes and DOIs checked on the Hub on 2026-10-20.
Versions
| Version | Released | Run | Where | Notes |
|---|---|---|---|---|
| Jachnun v1.0 current | 2026-10-19 | r6-en30 | Jachnun; also ff5aaa046a in the PoC record | 339.2 min; the first stable release |
| PoC-5 POC published as v5 | 2026-10-18 | r5-en30 | Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching @ 5522d9e86e | 294.2 min |
| PoC-4 POC published as v4 | 2026-10-18 | r4-en30 | Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching @ 2e02382e7d | 269.3 min |
| PoC-3 POC published as v3 | 2026-10-18 | r3-en30 | Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching @ 728e6f88d4 | 196.1 min; the MLX build is this version |
| PoC-2 POC published as v2 | 2026-10-15 | r2-en30 | Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching @ e1541065f8 | 120.1 min |
| PoC-1 POC published as v1 | 2026-10-15 | pilot-en30 | Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching @ ebc0e7aff1 | 77.4 min of code-switched audio |
Suggested implementations
Which build to use where. Every format is v1.0 except MLX. Full results and the training recipe: model card.
| Where | Runtime | Files |
|---|---|---|
| Android, Linux, Windows, iOS | sherpa-onnx, offline NeMo transducer, greedy search. What the Hebrish apps use; see Hebrish-App for working Kotlin and Python code | sherpa-onnx int8, ≈ 661 MB |
| Browser | parakeet.js on ONNX Runtime Web (WASM). How /demo works; about 2.3 GB of memory once loaded | ONNX int8 |
| Python | onnx-asr or ONNX Runtime directly | ONNX fp32 and int8 |
| Desktop and Android apps built on transcribe.cpp (Offline Voice Input, Handy) | transcribe.cpp v0.1.3 | GGUF Q8_0 (730 MB), Q4_K_M (475 MB) for phones with 4 GB of RAM or less |
| Mac (Apple Silicon) | pip install parakeet-mlx | MLX still PoC-3 |
| GPU, further fine-tuning | NVIDIA NeMo | .nemo, 2.47 GB |
Audio in is 16 kHz mono. Text out is English, with Hebrew words romanized as in the lexicon, never in Hebrew script.
Cite
Cite Jachnun with its DOI, 10.57967/hf/10995 (danielrosehill/Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching). To cite one format, use that repository's DOI from the table above.
@misc{rosehill_jachnun_2026,
author = {Daniel Rosehill},
title = {Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching},
year = {2026},
publisher = {Hugging Face},
url = {https://huggingface.co/danielrosehill/Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching},
doi = {10.57967/hf/10995}
}Request access
Request access on Hugging Face first. The Jachnun repositories are gated: open the repository, sign in, and fill in the short request form there. Requests are reviewed by hand. For non-commercial use, the attestation form covers what we ask. The Hebrish apps download this model from Hebrish's own CDN, cdn.hebrish.app, so using it in the apps needs no access request.
No answer after a few days? Request access here instead, or write to access-requests@hebrish.app. Commercial use, including telephony, needs a commercial licence.