List the XNNPACK int8 variants in config.json
Browse files- README.md +1 -1
- xnnpack/config.json +61 -0
README.md
CHANGED
|
@@ -46,7 +46,7 @@ tokenizer.json 2.4 MB
|
|
| 46 |
vulkan/config.json 2.6 kB
|
| 47 |
vulkan/whisper_small_vulkan_fp16.pte 881 MB
|
| 48 |
vulkan/whisper_small_vulkan_int8.pte 484 MB
|
| 49 |
-
xnnpack/config.json
|
| 50 |
xnnpack/whisper_small_xnnpack_fp32.pte 1.1 GB
|
| 51 |
xnnpack/whisper_small_xnnpack_int8.pte 421 MB
|
| 52 |
```
|
|
|
|
| 46 |
vulkan/config.json 2.6 kB
|
| 47 |
vulkan/whisper_small_vulkan_fp16.pte 881 MB
|
| 48 |
vulkan/whisper_small_vulkan_int8.pte 484 MB
|
| 49 |
+
xnnpack/config.json 2.6 kB
|
| 50 |
xnnpack/whisper_small_xnnpack_fp32.pte 1.1 GB
|
| 51 |
xnnpack/whisper_small_xnnpack_int8.pte 421 MB
|
| 52 |
```
|
xnnpack/config.json
CHANGED
|
@@ -68,6 +68,67 @@
|
|
| 68 |
]
|
| 69 |
}
|
| 70 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 71 |
}
|
| 72 |
]
|
| 73 |
}
|
|
|
|
| 68 |
]
|
| 69 |
}
|
| 70 |
}
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"file": "whisper_small_xnnpack_int8.pte",
|
| 74 |
+
"precision": "int8",
|
| 75 |
+
"methods": {
|
| 76 |
+
"decode": {
|
| 77 |
+
"inputs": [
|
| 78 |
+
{
|
| 79 |
+
"shape": [
|
| 80 |
+
1,
|
| 81 |
+
128
|
| 82 |
+
],
|
| 83 |
+
"dtype": "int64"
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"shape": [
|
| 87 |
+
128
|
| 88 |
+
],
|
| 89 |
+
"dtype": "int64"
|
| 90 |
+
},
|
| 91 |
+
{
|
| 92 |
+
"shape": [
|
| 93 |
+
1,
|
| 94 |
+
1500,
|
| 95 |
+
768
|
| 96 |
+
],
|
| 97 |
+
"dtype": "float32"
|
| 98 |
+
}
|
| 99 |
+
],
|
| 100 |
+
"outputs": [
|
| 101 |
+
{
|
| 102 |
+
"shape": [
|
| 103 |
+
1,
|
| 104 |
+
128,
|
| 105 |
+
51865
|
| 106 |
+
],
|
| 107 |
+
"dtype": "float32"
|
| 108 |
+
}
|
| 109 |
+
]
|
| 110 |
+
},
|
| 111 |
+
"encode": {
|
| 112 |
+
"inputs": [
|
| 113 |
+
{
|
| 114 |
+
"shape": [
|
| 115 |
+
480000
|
| 116 |
+
],
|
| 117 |
+
"dtype": "float32"
|
| 118 |
+
}
|
| 119 |
+
],
|
| 120 |
+
"outputs": [
|
| 121 |
+
{
|
| 122 |
+
"shape": [
|
| 123 |
+
1,
|
| 124 |
+
1500,
|
| 125 |
+
768
|
| 126 |
+
],
|
| 127 |
+
"dtype": "float32"
|
| 128 |
+
}
|
| 129 |
+
]
|
| 130 |
+
}
|
| 131 |
+
}
|
| 132 |
}
|
| 133 |
]
|
| 134 |
}
|