Sunno Whisper models

These are reproducible ONNX Runtime GenAI conversions used by Sunno, an offline live-captioning app.

Directory	Upstream model	Precision	Intended runtime
base/	openai/whisper-base	int8	ONNX Runtime GenAI on CPU
tiny/	openai/whisper-tiny	int8	ONNX Runtime GenAI on CPU

The files were generated directly from OpenAI's MIT-licensed Whisper weights with bench/convert_whisper_genai.py in the Sunno repository. They are not copies of a third-party ONNX conversion.

Each directory is a complete model and contains genai_config.json, the audio processor and tokenizer configuration, and the encoder and decoder ONNX graphs with their external data files.

Reproduce

python -m venv .venv-convert
.venv-convert\Scripts\python.exe -m pip install onnxruntime-genai onnx onnx_ir transformers torch
.venv-convert\Scripts\python.exe bench\convert_whisper_genai.py --model base tiny --out dist\onnx

The conversion targets the CPU execution provider because that is the native path on Windows ARM64. Sunno performs all inference locally after the initial model download.