shameez's picture
ONNX FP16 export of nvidia/nemotron-speech-streaming-en-0.6b (560ms latency)
d51d610 verified
Raw
History Blame Contribute Delete
2.17 kB
NVIDIA Nemotron Speech Streaming EN 0.6B — ONNX FP16 export (560ms latency)
=======================================================================
Licensed by NVIDIA Corporation under the NVIDIA Open Model License
This repository is an ONNX (FP16) export of the original NeMo checkpoint produced
directly from NVIDIA's official Hugging Face repository. It is a derivative work
distributed under the same license that governs the source model.
Source model
------------
Repository : nvidia/nemotron-speech-streaming-en-0.6b
URL : https://huggingface.co/nvidia/nemotron-speech-streaming-en-0.6b
Revision : 7a9b763e6c5fb103da690219c049fac917aa50b1
Conversion details
------------------
Export method : NeMo toolkit ASRModel.export() to ONNX
ONNX opset : 17
att_context_size: [70, 6] (560 ms streaming latency)
Precision : FP16 weights with FP32-typed graph I/O (Cast-wrapped).
The cache-aware pre_encode subsampling front-end is retained
in FP32 for converter compatibility; the 24-layer Conformer
body runs in FP16.
Chunk geometry : chunk_mel_frames = 56, pre_encode_cache_frames = 9,
total_input_frames = 65, chunk_audio_samples = 8960
License
-------
NVIDIA Open Model License Agreement (see LICENSE.txt for the full text).
URL: https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/
NOTE: This model is governed by the NVIDIA Open Model License.
It is NOT licensed under CC-BY-4.0 (a tag that has appeared on some
third-party exports of this checkpoint and is incorrect).
Toolkit versions used (full list in conversion/dep_baseline_nemo_export_env_macos.txt)
--------------------------------------------------------------------------------------
python 3.12.13
nemo_toolkit 2.7.3
torch 2.12.1+cpu
torchaudio 2.11.0+cpu
onnx 1.22.0
onnxruntime 1.27.0
onnxconverter-common 1.16.0
sentencepiece 0.2.1
numpy 1.26.4
Reproducibility
---------------
The export and conversion scripts (export_nemotron.py, convert_fp16.py) are
included in the conversion/ folder of this repository.