smpanaro's picture
Add new joint prefill+generation cache processor
07183f4
[
{
"metadataOutputVersion" : "3.0",
"outputSchema" : [
{
"hasShapeFlexibility" : "0",
"isOptional" : "0",
"dataType" : "Float16",
"formattedType" : "MultiArray (Float16 1 × 448 × 1 × 4096)",
"shortDescription" : "",
"shape" : "[1, 448, 1, 4096]",
"name" : "updated_k_cache",
"type" : "MultiArray"
},
{
"hasShapeFlexibility" : "0",
"isOptional" : "0",
"dataType" : "Float16",
"formattedType" : "MultiArray (Float16 1 × 4096 × 1 × 448)",
"shortDescription" : "",
"shape" : "[1, 4096, 1, 448]",
"name" : "updated_v_cache",
"type" : "MultiArray"
},
{
"hasShapeFlexibility" : "0",
"isOptional" : "0",
"dataType" : "Float16",
"formattedType" : "MultiArray (Float16)",
"shortDescription" : "",
"shape" : "[]",
"name" : "ignore_me_im_only_here_so_this_runs_on_the_ane",
"type" : "MultiArray"
}
],
"modelParameters" : [
],
"specificationVersion" : 7,
"mlProgramOperationTypeHistogram" : {
"SliceByIndex" : 2,
"Ios16.mul" : 1,
"Concat" : 2,
"Ios16.reduceMin" : 1
},
"computePrecision" : "Mixed (Float16, Int32)",
"isUpdatable" : "0",
"availability" : {
"macOS" : "13.0",
"tvOS" : "16.0",
"visionOS" : "1.0",
"watchOS" : "9.0",
"iOS" : "16.0",
"macCatalyst" : "16.0"
},
"modelType" : {
"name" : "MLModelType_mlProgram"
},
"userDefinedMetadata" : {
"com.github.apple.coremltools.source_dialect" : "TorchScript",
"com.github.apple.coremltools.source" : "torch==2.1.0",
"com.github.apple.coremltools.version" : "8.0b1"
},
"inputSchema" : [
{
"hasShapeFlexibility" : "0",
"isOptional" : "0",
"dataType" : "Float16",
"formattedType" : "MultiArray (Float16 1 × 448 × 1 × 4096)",
"shortDescription" : "",
"shape" : "[1, 448, 1, 4096]",
"name" : "old_k_cache",
"type" : "MultiArray"
},
{
"hasShapeFlexibility" : "0",
"isOptional" : "0",
"dataType" : "Float16",
"formattedType" : "MultiArray (Float16 1 × 64 × 1 × 4096)",
"shortDescription" : "",
"shape" : "[1, 64, 1, 4096]",
"name" : "new_k_cache",
"type" : "MultiArray"
},
{
"hasShapeFlexibility" : "0",
"isOptional" : "0",
"dataType" : "Float16",
"formattedType" : "MultiArray (Float16 1 × 4096 × 1 × 448)",
"shortDescription" : "",
"shape" : "[1, 4096, 1, 448]",
"name" : "old_v_cache",
"type" : "MultiArray"
},
{
"hasShapeFlexibility" : "0",
"isOptional" : "0",
"dataType" : "Float16",
"formattedType" : "MultiArray (Float16 1 × 4096 × 1 × 64)",
"shortDescription" : "",
"shape" : "[1, 4096, 1, 64]",
"name" : "new_v_cache",
"type" : "MultiArray"
}
],
"generatedClassName" : "cache_processor",
"method" : "predict"
}
]