[Vertex AI] Refactor HarmBlockThreshold as a struct and add .off (#13863)

Author: andrewheard

FirebaseVertexAI/CHANGELOG.md

@@ -46,6 +46,12 @@
 - [changed] The response from `GenerativeModel.countTokens(...)` now includes
   `systemInstruction`, `tools` and `generationConfig` in the `totalTokens` and
   `totalBillableCharacters` counts, where applicable. (#13813)
+- [added] Added a new `HarmCategory` `.civicIntegrity` for filtering content
+  that may be used to harm civic integrity. (#13728)
+- [added] Added a new `HarmBlockThreshold` `.off`, which turns off the safety
+  filter. (#13863)
+- [added] Added new `FinishReason` values `.blocklist`, `.prohibitedContent`,
+  `.spii` and `.malformedFunctionCall` that may be reported. (#13860)
 
 # 11.3.0
 - [added] Added `Decodable` conformance for `FunctionResponse`. (#13606)

FirebaseVertexAI/Sources/Safety.swift

@@ -90,18 +90,41 @@ public struct SafetyRating: Equatable, Hashable, Sendable {
 @available(iOS 15.0, macOS 11.0, macCatalyst 15.0, tvOS 15.0, watchOS 8.0, *)
 public struct SafetySetting {
   /// Block at and beyond a specified ``SafetyRating/HarmProbability``.
-  public enum HarmBlockThreshold: String, Sendable {
-    // Content with `.negligible` will be allowed.
-    case blockLowAndAbove = "BLOCK_LOW_AND_ABOVE"
+  public struct HarmBlockThreshold: EncodableProtoEnum, Sendable {
+    enum Kind: String {
+      case blockLowAndAbove = "BLOCK_LOW_AND_ABOVE"
+      case blockMediumAndAbove = "BLOCK_MEDIUM_AND_ABOVE"
+      case blockOnlyHigh = "BLOCK_ONLY_HIGH"
+      case blockNone = "BLOCK_NONE"
+      case off = "OFF"
+    }
+
+    /// Content with `.negligible` will be allowed.
+    public static var blockLowAndAbove: HarmBlockThreshold {
+      return self.init(kind: .blockLowAndAbove)
+    }
 
     /// Content with `.negligible` and `.low` will be allowed.
-    case blockMediumAndAbove = "BLOCK_MEDIUM_AND_ABOVE"
+    public static var blockMediumAndAbove: HarmBlockThreshold {
+      return self.init(kind: .blockMediumAndAbove)
+    }
 
     /// Content with `.negligible`, `.low`, and `.medium` will be allowed.
-    case blockOnlyHigh = "BLOCK_ONLY_HIGH"
+    public static var blockOnlyHigh: HarmBlockThreshold {
+      return self.init(kind: .blockOnlyHigh)
+    }
 
     /// All content will be allowed.
-    case blockNone = "BLOCK_NONE"
+    public static var blockNone: HarmBlockThreshold {
+      return self.init(kind: .blockNone)
+    }
+
+    /// Turn off the safety filter.
+    public static var off: HarmBlockThreshold {
+      return self.init(kind: .off)
+    }
+
+    let rawValue: String
   }
 
   enum CodingKeys: String, CodingKey {

FirebaseVertexAI/Tests/Integration/IntegrationTests.swift

@@ -84,6 +84,18 @@ final class IntegrationTests: XCTestCase {
 
   func testCountTokens_text() async throws {
     let prompt = "Why is the sky blue?"
+    model = vertex.generativeModel(
+      modelName: "gemini-1.5-pro",
+      generationConfig: generationConfig,
+      safetySettings: [
+        SafetySetting(harmCategory: .harassment, threshold: .blockLowAndAbove),
+        SafetySetting(harmCategory: .hateSpeech, threshold: .blockMediumAndAbove),
+        SafetySetting(harmCategory: .sexuallyExplicit, threshold: .blockOnlyHigh),
+        SafetySetting(harmCategory: .dangerousContent, threshold: .blockNone),
+        SafetySetting(harmCategory: .civicIntegrity, threshold: .off),
+      ],
+      systemInstruction: systemInstruction
+    )
 
     let response = try await model.countTokens(prompt)