johnpattern/safenotes-funasr-nano-ov

Fun-ASR-Nano-2512 (OpenVINO int8) for Safenotes

FunAudioLLM / Tongyi Lab's Fun-ASR-Nano-2512 speech-recognition model (strong on Chinese) converted to OpenVINO IR with int8 weights.

This is a format conversion made by Safenotes (a local, on-device meeting notetaker) so the model runs on Intel hardware (OpenVINO). It is not an official release of the original authors, who do not endorse it. All credit for the model goes to them.

Original model and attribution

• Fun-ASR-Nano-2512 by FunAudioLLM / Tongyi Lab (Alibaba): FunAudioLLM/Fun-ASR-Nano-2512 at revision 272c57b82523ada6fd87095e955f8e29100979ab, licensed Apache-2.0 (model card metadata; the repo has no LICENSE file). Its LLM decoder is Qwen3-0.6B (Qwen team, Alibaba Cloud, Apache-2.0); the tokenizer files (vocab.json, merges.txt, tokenizer.json, …) are Qwen3's, as shipped in the upstream repo.

This repository is distributed under the same licence: Apache License 2.0. The full licence text is in LICENSE.

Changes made by Safenotes

model.pt (sha256 55ae0d2fee36…) exported with optimum-intel 2.2's FunASR export (funasr 1.4.19, MIT) and int8 weight compression (NNCF); the tokenizer converted to OpenVINO tokenizer/detokenizer IRs. No retraining.

How Safenotes uses it

Safenotes uses it for Chinese transcripts after the meeting on PCs without a usable GPU. The Safenotes app downloads the files below from a pinned revision of this repository and checks every file's sha256 before using it. Everything runs on the user's PC; no audio or text leaves it.

Files

16 files, 863,684,371 bytes.

file	bytes	sha256
added_tokens.json	707	c0284b582e14987fbd3d5a2cb2bd139084371ed9acbae488829a1c900833c680
chat_template.jinja	4,168	a55ee1b1660128b7098723e0abcd92caa0788061051c62d51cbe87d9cf1974d8
config.json	471	4101a8a999b9a17ad2edac78387bac4e7bc2560a4580b77f87fdae9d9d2aa893
merges.txt	1,671,853	8831e4f1a044471340f7c0a83d7bd71306a5b867e95fd870f74d0c5308a904d5
openvino_decoder_model.bin	597,603,872	5db73a3595036fa77d39145f233cc106dc41c9b66bbb66bc3038d3b3a58a1536
openvino_decoder_model.xml	3,029,667	2772d74bdc46ff7d6cde97f32918952dd0673b4b2edd90a5dba5a205d75f2d76
openvino_detokenizer.bin	2,189,717	d73093f72c80ebcaffb76a192727170ecd3c6a06c4df52df0e76b3f70ad9d7f7
openvino_detokenizer.xml	11,725	e9fcfd7037de734fb3da4897dd8de8b4a262029c004deb3eecce8081ab722b78
openvino_encoder_model.bin	235,349,042	8d7a0c124b78a5fbfdcad5979ef9be64497d71500398f328b16776354c17750e
openvino_encoder_model.xml	3,998,654	0f0042a7d905fe2aacdcfc0b1201cb2608025f8f6b1c19a143f0bcc9bd756f59
openvino_tokenizer.bin	5,588,841	1ed7ef1d43756fa0f198acbf061b36d3548d6be51eebc851130c2868b9d81d54
openvino_tokenizer.xml	30,150	533b51a58c44a9946501e219117852a7ba10d73749e0c2c514a68e274d73b3a6
special_tokens_map.json	613	76862e765266b85aa9459767e33cbaf13970f327a0e88d1c65846c2ddd3a1ecd
tokenizer.json	11,422,654	aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
tokenizer_config.json	5,404	443bfa629eb16387a12edbf92a76f6a6f10b2af3b53d87ba1550adfcf45f7fa0
vocab.json	2,776,833	ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910

No warranty

Provided "as is", without warranties or conditions of any kind, express or implied, including fitness for a particular purpose and accuracy of the output (see the licence). Use at your own risk.