Drop unused joint_decision.onnx.data — only int8 (or fp32 for VAD/wespeaker) is referenced by the manifest

Drop unused joint_decision.onnx — only int8 (or fp32 for VAD/wespeaker) is referenced by the manifest

Drop unused decoder.onnx — only int8 (or fp32 for VAD/wespeaker) is referenced by the manifest

Drop unused encoder.onnx.data — only int8 (or fp32 for VAD/wespeaker) is referenced by the manifest

Drop unused encoder.onnx — only int8 (or fp32 for VAD/wespeaker) is referenced by the manifest

Drop unused preprocessor.onnx.data — only int8 (or fp32 for VAD/wespeaker) is referenced by the manifest

Drop unused preprocessor.onnx — only int8 (or fp32 for VAD/wespeaker) is referenced by the manifest

Remove stale decoder.onnx.data — decoder.onnx is now self-contained

Re-export with dynamo=False + QUInt8 quant

Re-export with joint input names encoder_step / decoder_step matching CoreML contract (#1290)

Upload config.json with huggingface_hub

Upload vocab.json with huggingface_hub

Upload joint_decision.int8.onnx with huggingface_hub

Upload decoder.int8.onnx with huggingface_hub

Upload encoder.int8.onnx with huggingface_hub

Upload preprocessor.int8.onnx with huggingface_hub

Upload README.md with huggingface_hub

initial commit