msluszniak's picture
Export the LR-ASPP QNN variants as a8w8 (2.4x faster at the same mIoU)
db9deda verified
Raw History Blame Contribute Delete
3.4 kB
{
"$schema": "https://huggingface.co/software-mansion/react-native-executorch-spec/resolve/main/config.schema.json",
"model": "lraspp_mobilenet_v3_large",
"family": "lraspp",
"capabilities": [
"semantic-segmentation"
],
"backend": "qnn",
"license": "bsd-3-clause",
"variants": [
{
"file": "lraspp_mobilenet_v3_large_qnn_a8w8_v69.pte",
"precision": "a8w8",
"quantized": true,
"default": false,
"methods": {
"forward": {
"inputs": [
{
"shape": [
1,
3,
520,
520
],
"dtype": "float32"
}
],
"outputs": [
{
"shape": [
1,
520,
520
],
"dtype": "int32"
}
]
}
}
},
{
"file": "lraspp_mobilenet_v3_large_qnn_a8w8_v73.pte",
"precision": "a8w8",
"quantized": true,
"default": false,
"methods": {
"forward": {
"inputs": [
{
"shape": [
1,
3,
520,
520
],
"dtype": "float32"
}
],
"outputs": [
{
"shape": [
1,
520,
520
],
"dtype": "int32"
}
]
}
}
},
{
"file": "lraspp_mobilenet_v3_large_qnn_a8w8_v75.pte",
"precision": "a8w8",
"quantized": true,
"default": false,
"methods": {
"forward": {
"inputs": [
{
"shape": [
1,
3,
520,
520
],
"dtype": "float32"
}
],
"outputs": [
{
"shape": [
1,
520,
520
],
"dtype": "int32"
}
]
}
}
},
{
"file": "lraspp_mobilenet_v3_large_qnn_a8w8_v79.pte",
"precision": "a8w8",
"quantized": true,
"default": false,
"methods": {
"forward": {
"inputs": [
{
"shape": [
1,
3,
520,
520
],
"dtype": "float32"
}
],
"outputs": [
{
"shape": [
1,
520,
520
],
"dtype": "int32"
}
]
}
}
},
{
"file": "lraspp_mobilenet_v3_large_qnn_a8w8_v81.pte",
"precision": "a8w8",
"quantized": true,
"default": true,
"methods": {
"forward": {
"inputs": [
{
"shape": [
1,
3,
520,
520
],
"dtype": "float32"
}
],
"outputs": [
{
"shape": [
1,
520,
520
],
"dtype": "int32"
}
]
}
}
}
]
}