ONNX
webgpu
int4
ASMR-onnx / tts_int4 /quantization.json
GCStream's picture
Quantized ONNX graphs for browser inference
319d2ff verified
Raw History Blame Contribute Delete
830 Bytes
{
"mode": "tts",
"bits": 4,
"block": 64,
"symmetric": true,
"graphs": [
{
"src": "dit_step.onnx",
"dst": "dit_step.onnx",
"src_mb": 654.0,
"dst_mb": 388.2,
"seconds": 2.3,
"kind": "dit",
"verify": {
"max_abs": 1.18701171875,
"rel": 0.17334569990634918,
"mean_rel": 0.15340197086334229,
"corr": 0.987980546554555
}
},
{
"src": "vocoder.onnx",
"dst": "vocoder.onnx",
"src_mb": 26.0,
"dst_mb": 7.6,
"seconds": 0.1,
"kind": "vocoder",
"verify": {
"max_abs": 7.4287109375,
"rel": 0.17806647717952728,
"mean_rel": 0.11186939477920532,
"corr": 0.9849824359828151
}
}
],
"total": {
"src_mb": 680.0,
"dst_mb": 395.8,
"ratio": 0.582
}
}