upload prefill_t32.mlmodelc/metadata.json
Browse files
prefill_t32.mlmodelc/metadata.json
ADDED
|
@@ -0,0 +1,135 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"metadataOutputVersion" : "3.0",
|
| 4 |
+
"storagePrecision" : "Mixed (Float16, Int8)",
|
| 5 |
+
"outputSchema" : [
|
| 6 |
+
{
|
| 7 |
+
"hasShapeFlexibility" : "0",
|
| 8 |
+
"isOptional" : "0",
|
| 9 |
+
"dataType" : "Int32",
|
| 10 |
+
"formattedType" : "MultiArray (Int32 1)",
|
| 11 |
+
"shortDescription" : "",
|
| 12 |
+
"shape" : "[1]",
|
| 13 |
+
"name" : "token_id",
|
| 14 |
+
"type" : "MultiArray"
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"hasShapeFlexibility" : "0",
|
| 18 |
+
"isOptional" : "0",
|
| 19 |
+
"dataType" : "Float16",
|
| 20 |
+
"formattedType" : "MultiArray (Float16 1)",
|
| 21 |
+
"shortDescription" : "",
|
| 22 |
+
"shape" : "[1]",
|
| 23 |
+
"name" : "token_logit",
|
| 24 |
+
"type" : "MultiArray"
|
| 25 |
+
}
|
| 26 |
+
],
|
| 27 |
+
"modelParameters" : [
|
| 28 |
+
|
| 29 |
+
],
|
| 30 |
+
"specificationVersion" : 10,
|
| 31 |
+
"mlProgramOperationTypeHistogram" : {
|
| 32 |
+
"Ios18.softmax" : 18,
|
| 33 |
+
"Ios19.mul" : 399,
|
| 34 |
+
"Ios18.matmul" : 72,
|
| 35 |
+
"Ios19.sliceUpdate" : 36,
|
| 36 |
+
"Ios18.gatherAlongAxis" : 1,
|
| 37 |
+
"Ios19.squeeze" : 56,
|
| 38 |
+
"Ios18.readState" : 36,
|
| 39 |
+
"Tile" : 36,
|
| 40 |
+
"Ios18.gather" : 5,
|
| 41 |
+
"Ios16.reduceSum" : 1,
|
| 42 |
+
"Ios19.add" : 128,
|
| 43 |
+
"Ios18.layerNorm" : 109,
|
| 44 |
+
"Ios18.writeState" : 36,
|
| 45 |
+
"Ios19.concat" : 145,
|
| 46 |
+
"Ios19.constexprBlockwiseShiftScale" : 132,
|
| 47 |
+
"Ios19.transpose" : 146,
|
| 48 |
+
"Ios18.reduceArgmax" : 1,
|
| 49 |
+
"Ios19.expandDims" : 117,
|
| 50 |
+
"Ios18.conv" : 127,
|
| 51 |
+
"Ios18.cast" : 1,
|
| 52 |
+
"Ios18.gelu" : 18,
|
| 53 |
+
"Ios19.split" : 145,
|
| 54 |
+
"Ios19.sliceByIndex" : 37,
|
| 55 |
+
"Ios19.sub" : 1,
|
| 56 |
+
"Ios19.select" : 2,
|
| 57 |
+
"Ios19.greaterEqual" : 2,
|
| 58 |
+
"Ios19.reshape" : 108
|
| 59 |
+
},
|
| 60 |
+
"computePrecision" : "Mixed (Float16, Int32, UInt16)",
|
| 61 |
+
"isUpdatable" : "0",
|
| 62 |
+
"stateSchema" : [
|
| 63 |
+
{
|
| 64 |
+
"dataType" : "Float16",
|
| 65 |
+
"isOptional" : "0",
|
| 66 |
+
"formattedType" : "State (Float16 36 × 1 × 2048 × 256)",
|
| 67 |
+
"shortDescription" : "",
|
| 68 |
+
"shape" : "[36, 1, 2048, 256]",
|
| 69 |
+
"name" : "kv_cache_0",
|
| 70 |
+
"type" : "State"
|
| 71 |
+
}
|
| 72 |
+
],
|
| 73 |
+
"availability" : {
|
| 74 |
+
"macOS" : "16.0",
|
| 75 |
+
"tvOS" : "19.0",
|
| 76 |
+
"visionOS" : "3.0",
|
| 77 |
+
"watchOS" : "12.0",
|
| 78 |
+
"iOS" : "19.0",
|
| 79 |
+
"macCatalyst" : "19.0"
|
| 80 |
+
},
|
| 81 |
+
"modelType" : {
|
| 82 |
+
"name" : "MLModelType_mlProgram"
|
| 83 |
+
},
|
| 84 |
+
"userDefinedMetadata" : {
|
| 85 |
+
"com.github.apple.coremltools.conversion_date" : "2026-04-23",
|
| 86 |
+
"com.github.apple.coremltools.source" : "torch==2.8.0",
|
| 87 |
+
"com.github.apple.coremltools.version" : "9.0",
|
| 88 |
+
"com.github.apple.coremltools.source_dialect" : "TorchScript"
|
| 89 |
+
},
|
| 90 |
+
"inputSchema" : [
|
| 91 |
+
{
|
| 92 |
+
"hasShapeFlexibility" : "0",
|
| 93 |
+
"isOptional" : "0",
|
| 94 |
+
"dataType" : "Int32",
|
| 95 |
+
"formattedType" : "MultiArray (Int32 1 × 32)",
|
| 96 |
+
"shortDescription" : "",
|
| 97 |
+
"shape" : "[1, 32]",
|
| 98 |
+
"name" : "input_ids",
|
| 99 |
+
"type" : "MultiArray"
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"hasShapeFlexibility" : "0",
|
| 103 |
+
"isOptional" : "0",
|
| 104 |
+
"dataType" : "Int32",
|
| 105 |
+
"formattedType" : "MultiArray (Int32 32)",
|
| 106 |
+
"shortDescription" : "",
|
| 107 |
+
"shape" : "[32]",
|
| 108 |
+
"name" : "position_ids",
|
| 109 |
+
"type" : "MultiArray"
|
| 110 |
+
},
|
| 111 |
+
{
|
| 112 |
+
"hasShapeFlexibility" : "0",
|
| 113 |
+
"isOptional" : "0",
|
| 114 |
+
"dataType" : "Float16",
|
| 115 |
+
"formattedType" : "MultiArray (Float16 1 × 1 × 32 × 2048)",
|
| 116 |
+
"shortDescription" : "",
|
| 117 |
+
"shape" : "[1, 1, 32, 2048]",
|
| 118 |
+
"name" : "causal_mask",
|
| 119 |
+
"type" : "MultiArray"
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"hasShapeFlexibility" : "0",
|
| 123 |
+
"isOptional" : "0",
|
| 124 |
+
"dataType" : "Float16",
|
| 125 |
+
"formattedType" : "MultiArray (Float16 1 × 1 × 2048 × 32)",
|
| 126 |
+
"shortDescription" : "",
|
| 127 |
+
"shape" : "[1, 1, 2048, 32]",
|
| 128 |
+
"name" : "update_mask",
|
| 129 |
+
"type" : "MultiArray"
|
| 130 |
+
}
|
| 131 |
+
],
|
| 132 |
+
"generatedClassName" : "prefill_t32",
|
| 133 |
+
"method" : "predict"
|
| 134 |
+
}
|
| 135 |
+
]
|