Declare the int8 Core ML variant
Browse files- coreml/config.json +38 -0
coreml/config.json
CHANGED
|
@@ -8,9 +8,47 @@
|
|
| 8 |
"backend": "coreml",
|
| 9 |
"license": "apache-2.0",
|
| 10 |
"variants": [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
{
|
| 12 |
"file": "all_minilm_l6_v2_coreml_fp16.pte",
|
| 13 |
"precision": "fp16",
|
|
|
|
|
|
|
| 14 |
"methods": {
|
| 15 |
"forward": {
|
| 16 |
"inputs": [
|
|
|
|
| 8 |
"backend": "coreml",
|
| 9 |
"license": "apache-2.0",
|
| 10 |
"variants": [
|
| 11 |
+
{
|
| 12 |
+
"file": "all_minilm_l6_v2_coreml_int8.pte",
|
| 13 |
+
"precision": "int8",
|
| 14 |
+
"quantized": true,
|
| 15 |
+
"default": true,
|
| 16 |
+
"methods": {
|
| 17 |
+
"forward": {
|
| 18 |
+
"inputs": [
|
| 19 |
+
{
|
| 20 |
+
"shape": [
|
| 21 |
+
1,
|
| 22 |
+
254
|
| 23 |
+
],
|
| 24 |
+
"dtype": "int64"
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"shape": [
|
| 28 |
+
1,
|
| 29 |
+
254
|
| 30 |
+
],
|
| 31 |
+
"dtype": "int64"
|
| 32 |
+
}
|
| 33 |
+
],
|
| 34 |
+
"outputs": [
|
| 35 |
+
{
|
| 36 |
+
"shape": [
|
| 37 |
+
1,
|
| 38 |
+
384
|
| 39 |
+
],
|
| 40 |
+
"dtype": "float32"
|
| 41 |
+
}
|
| 42 |
+
]
|
| 43 |
+
}
|
| 44 |
+
},
|
| 45 |
+
"quantize": "int8"
|
| 46 |
+
},
|
| 47 |
{
|
| 48 |
"file": "all_minilm_l6_v2_coreml_fp16.pte",
|
| 49 |
"precision": "fp16",
|
| 50 |
+
"quantized": false,
|
| 51 |
+
"default": true,
|
| 52 |
"methods": {
|
| 53 |
"forward": {
|
| 54 |
"inputs": [
|