Take SwiftProtobuf dependency out from ModelSupport (#694)

Author: xihui-wu

Models/Text/BERT/BERT.swift

@@ -16,8 +16,22 @@ import Checkpoints
 import Datasets
 import Foundation
 import ModelSupport
+import SwiftProtobuf
 import TensorFlow
 
+extension Vocabulary {
+    public init(fromSentencePieceModel fileURL: URL) throws {
+        self.init(
+            tokensToIds: [String: Int](
+                (try Sentencepiece_ModelProto(serializedData: Data(contentsOf: fileURL)))
+                    .pieces
+                    .map { $0.piece.replacingOccurrences(of: "▁", with: "##") }
+                    .map { $0 == "<unk>" ? "[UNK]" : $0 }
+                    .enumerated().map { ($0.element, $0.offset) },
+                uniquingKeysWith: { (v1, v2) in max(v1, v2) }))
+    }
+}
+
 /// Represents a type that can contribute to the regularization term when training models.
 public protocol Regularizable: Differentiable {
     /// The contribution of this term to the regularization term. This should be set to

Models/Text/CMakeLists.txt

@@ -10,6 +10,7 @@ add_library(TextModels
   GPT2/TransformerLM.swift
   GPT2/Operators.swift
   GPT2/PythonCheckpointReader.swift
+  SentencePiece/SentencePieceModel.pb.swift
   WordSeg/Lattice.swift
   WordSeg/Model.swift
   WordSeg/SemiRing.swift)
@@ -18,7 +19,8 @@ set_target_properties(TextModels PROPERTIES
 target_compile_options(TextModels PRIVATE
   $<$<BOOL:${BUILD_TESTING}>:-enable-testing>)
 target_link_libraries(TextModels PUBLIC
-  Datasets)
+  Datasets
+  SwiftProtobuf)
 
 install(TARGETS TextModels
   ARCHIVE DESTINATION lib/swift/$<LOWER_CASE:${CMAKE_SYSTEM_NAME}>

Support/Text/SentencePiece/SentencePieceModel.pb.swift renamed to Models/Text/SentencePiece/SentencePieceModel.pb.swift

@@ -33,11 +33,11 @@ let package = Package(
         .target(name: "Datasets", dependencies: ["ModelSupport"], path: "Datasets"),
         .target(name: "STBImage", path: "Support/STBImage"),
         .target(
-            name: "ModelSupport", dependencies: ["SwiftProtobuf", "STBImage"], path: "Support",
+            name: "ModelSupport", dependencies: ["STBImage"], path: "Support",
             exclude: ["STBImage"]),
         .target(name: "ImageClassificationModels", path: "Models/ImageClassification"),
         .target(name: "VideoClassificationModels", path: "Models/Spatiotemporal"),
-        .target(name: "TextModels", dependencies: ["Checkpoints", "Datasets"], path: "Models/Text"),
+        .target(name: "TextModels", dependencies: ["Checkpoints", "Datasets", "SwiftProtobuf"], path: "Models/Text"),
         .target(name: "RecommendationModels", path: "Models/Recommendation"),
         .target(name: "TrainingLoop", dependencies: ["ModelSupport"], path: "TrainingLoop"),
         .target(

Support/Text/SentencePiece/SentencePieceModel.proto renamed to Models/Text/SentencePiece/SentencePieceModel.proto

@@ -13,7 +13,6 @@ add_library(ModelSupport
   AnyLayer.swift
   AnyLayerTangentVector.swift
   Text/BytePairEncoder.swift
-  Text/SentencePiece/SentencePieceModel.pb.swift
   Text/TextBatch.swift
   Text/Tokenization.swift
   Text/WordSeg/Alphabet.swift

Package.swift

@@ -88,17 +88,6 @@ extension Vocabulary {
 }
 
 extension Vocabulary {
-    public init(fromSentencePieceModel fileURL: URL) throws {
-        self.init(
-            tokensToIds: [String: Int](
-                (try Sentencepiece_ModelProto(serializedData: Data(contentsOf: fileURL)))
-                    .pieces
-                    .map { $0.piece.replacingOccurrences(of: "▁", with: "##") }
-                    .map { $0 == "<unk>" ? "[UNK]" : $0 }
-                    .enumerated().map { ($0.element, $0.offset) },
-                uniquingKeysWith: { (v1, v2) in max(v1, v2) }))
-    }
-
     public init(fromJSONFile fileURL: URL) throws {
         let json = try String(contentsOfFile: fileURL.path)
         let tokensToIds = try JSONDecoder().decode(