---
license: other
base_model: nvidia/nemotron-3.5-asr-streaming-0.6b
tags:
- onnx
- asr
- streaming
- nemotron
- nemo
- sherpa-onnx
- multilingual
---

# Streaming Nemotron 3.5 ASR (0.6B, multilingual) — 560 ms — sherpa-onnx ONNX

sherpa-onnx streaming transducer export of
[nvidia/nemotron-3.5-asr-streaming-0.6b](https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b)
at the **560 ms** cache-aware chunk (`att_context_size = [56, 6]`), produced with
sherpa-onnx's official `scripts/nemo/nemotron-3.5-asr-streaming-0.6b/export_onnx.py`.

Ships the full package:
- fp32: `encoder.onnx` + `encoder.data`, `decoder.onnx`, `joiner.onnx`
- int8: `encoder.int8.onnx`, `decoder.int8.onnx`, `joiner.int8.onnx` (ORT dynamic)
- fp16: `encoder.fp16.onnx`, `decoder.fp16.onnx`, `joiner.fp16.onnx` (onnxruntime float16, `keep_io_types`)
- `tokens.txt`

The multilingual `prompt_index` encoder input is exposed for per-stream language selection
(or `auto` = the metadata `auto_prompt_id`). Sibling of the `-1120ms` export.

Made for [WinSTT](https://github.com/dahshury/WinSTT).