Nemotron 3.5 ASR Streaming 0.6B — ONNX FP32 (CPU)
Full-precision FP32 ONNX model converted from NVIDIA Nemotron 3.5 ASR Streaming 0.6B using Microsoft Olive.
Maximum quality for CPU inference with ONNX Runtime GenAI. No quantization — pure FP32 weights.
Model Details
| Property | Value |
|---|---|
| Base model | NVIDIA Nemotron 3.5 ASR Streaming 0.6B |
| Encoder layers | 24 |
| Hidden size | 1024 |
| Precision | FP32 (no quantization) |
| Chunk size | 0.56s (8,960 samples @ 16kHz) |
| Sample rate | 16,000 Hz |
| Vocab size | 13,088 |
| Languages | 80+ (multilingual) |
| VAD | Silero VAD included |
| Total size | ~2,479 MB |
Quick Start
C# / .NET
using NemotronSpeech;
using var session = new ModelSession(
modelPath: "DimQ1/nemotron-3.5-asr-streaming-0.6b-onnx-fp32-cpu",
executionProvider: "cpu",
langId: "11", // Russian
useVad: true
);
Python
import onnxruntime_genai as og
model = og.Model("DimQ1/nemotron-3.5-asr-streaming-0.6b-onnx-fp32-cpu")
processor = og.StreamingProcessor(model)
# ...
Files
| File | Size |
|---|---|
encoder.onnx + .data | 2,380 MB |
decoder.onnx + .data | 57 MB |
joint.onnx + .data | 36 MB |
silero_vad.onnx | 2.1 MB |
tokenizer.json | 0.7 MB |
Available Variants
| Model | Precision | Encoder | Total | Quality |
|---|---|---|---|---|
| FP32 | FP32 | 2,380 MB | 2,479 MB | ⭐ Best |
| INT8 | INT8 k-quant | 922 MB | 1,021 MB | Good |
| INT4 | INT4 k-quant | 658 MB | 757 MB | ⚡ Fastest |
License
Inherits cc-by-nc-4.0 from NVIDIA.