nemotron-3.5-asr-streaming-0.6b-6bit-coreml

nvidia/nemotron-3.5-asr-streaming-0.6b for Apple silicon: a transcribe model package of compiled Core ML models, mostly on the Neural Engine. The transcribe-model-darwin Rust crate loads it (AnyModel::load).

Package	nemotron-3.5-asr-0.6b: format 3, architecture nemotron-rnnt
Source weights	https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b/resolve/main/nemotron-3.5-asr-streaming-0.6b.nemo
License	LicenseRef-OpenMDW-1.1 (LICENSE, NOTICE)
Languages	28: ar, bg, cs, da, de, en, es, et, fi, fr, hi, hr, hu, it, ja, ko, nb, nl, pl, pt, ro, ru, sk, sv, tr, uk, vi, zh
Size	449 MiB in 14 files
Requires	macOS 15.0 on Apple silicon
Built with	transcribe-models 0.0.1 at commit ff6f17639ee4d20f3724f9dfa6d99c34755dc77b, coremltools 9.0

Models

Model	Compute units	Functions	Weights
frontend	CPU	window_14_chunk_14	not compressed
encoder	CPU and Neural Engine	window_14_chunk_14	6-bit k-means palettes per tensor

Usage

Download manifest.json and every file it lists from https://huggingface.co/yorganci/nemotron-3.5-asr-streaming-0.6b-6bit-coreml/resolve/<revision>/<path>, checking each file's size and SHA-256 against the manifest, then load the directory:

use transcribe_core::{Audio, TranscribeOptions, Transcriber};
use transcribe_model_darwin::{AnyModel, LoadOptions};

let mut model = AnyModel::load("nemotron-3.5-asr-0.6b", LoadOptions::default())?;
let transcript = model.transcribe(Audio::new(&samples, 16_000), &TranscribeOptions::default())?;

The first load on a machine compiles the models for the Neural Engine (seconds to a couple of minutes); later loads are fast.

Evaluation

Error rates of this build with transcribe-eval (Apple M4; characters for Japanese, words otherwise):

Set	Error rate	Substitutions	Deletions	Insertions	Reference units
LibriSpeech test-clean, 20 utterances	2.40 %	9	0	3	500
LibriSpeech test-clean, 62 s clip, long-form with timestamps	1.41 %	2	0	0	142
FLEURS de, 10 utterances	12.32 %	19	5	2	211
FLEURS fr, 10 utterances	17.06 %	16	5	22	252
FLEURS es, 10 utterances	9.38 %	12	4	8	256
FLEURS ja, 10 utterances	11.47 %	38	7	8	462

License

nemotron-3.5-asr-0.6b

NVIDIA Nemotron 3.5 ASR Streaming 0.6B (https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b) by NVIDIA, licensed under the OpenMDW License Agreement, version 1.1.

Source: https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b/resolve/main/nemotron-3.5-asr-streaming-0.6b.nemo
License: LicenseRef-OpenMDW-1.1 (LICENSE)

Changes: this package is a modified form of the source weights. transcribe-models
(https://github.com/atahanyorganci/transcribe) ported the model to PyTorch and converted
it to Core ML with coremltools 9.0, in float16 (the front end's constants in float32),
with these weight compressions:

- frontend: not compressed
- encoder: 6-bit k-means palettes per tensor
- decoder (decoder.safetensors, read by the runtime): the prediction network and the joint in int8 per row