Use seconds instead of milliseconds for latency metrics.

PiperOrigin-RevId: 687349834
google-ai-edge · Oct 18, 2024 · a460350 · a460350
1 parent f892e20
commit a460350
Show file tree

Hide file tree

Showing 2 changed files with 13 additions and 16 deletions.
diff --git a/mediapipe/tasks/ios/genai/inference/sources/LlmInference+Session.swift b/mediapipe/tasks/ios/genai/inference/sources/LlmInference+Session.swift
@@ -67,7 +67,7 @@ extension LlmInference {
         } ?? llmInference.createSessionRunner(sessionConfig: sessionConfig)
 
       self.llmInference = llmInference
-      self.metrics = Metrics(responseGenerationTimeInMillis: 0)
+      self.metrics = Metrics(responseGenerationTimeInSeconds: 0)
       super.init()
     }
 
@@ -97,7 +97,7 @@ extension LlmInference {
     init(llmSessionRunner: LlmSessionRunner, llmInference: LlmInference) {
       self.llmSessionRunner = llmSessionRunner
       self.llmInference = llmInference
-      self.metrics = Metrics(responseGenerationTimeInMillis: 0)
+      self.metrics = Metrics(responseGenerationTimeInSeconds: 0)
       super.init()
     }
 
@@ -256,9 +256,8 @@ extension LlmInference {
     private func markResponseGenerationCompleted() {
       llmInference.markResponseGenerationCompleted()
       metrics = Metrics(
-        responseGenerationTimeInMillis: (TimeInterval(clock_gettime_nsec_np(CLOCK_MONOTONIC_RAW))
-          - responseGenerationStartTime) * 1000
-          / TimeInterval(NSEC_PER_SEC))
+        responseGenerationTimeInSeconds: (TimeInterval(clock_gettime_nsec_np(CLOCK_MONOTONIC_RAW))
+          - responseGenerationStartTime) / TimeInterval(NSEC_PER_SEC))
     }
 
     private static func humanReadableString(
@@ -310,13 +309,11 @@ extension LlmInference.Session {
   /// Note: Inherits from `NSObject` for Objective C interoperability.
   @objc(MPPLLMInferenceSessionMetrics) public final class Metrics: NSObject {
 
-    /// The time it took to generate the full response for last query, in milliseconds.
-    @objc public private(set) var responseGenerationTimeInMillis: TimeInterval
+    /// The time it took to generate the full response for last query, in seconds.
+    @objc public private(set) var responseGenerationTimeInSeconds: TimeInterval
 
-    @objc public init(
-      responseGenerationTimeInMillis: TimeInterval
-    ) {
-      self.responseGenerationTimeInMillis = responseGenerationTimeInMillis
+    @objc public init(responseGenerationTimeInSeconds: TimeInterval) {
+      self.responseGenerationTimeInSeconds = responseGenerationTimeInSeconds
     }
   }
 }

diff --git a/mediapipe/tasks/ios/genai/inference/sources/LlmInference.swift b/mediapipe/tasks/ios/genai/inference/sources/LlmInference.swift
@@ -81,7 +81,7 @@ import MediaPipeTasksGenAIC
     }
     let timeAfterInit = clock_gettime_nsec_np(CLOCK_MONOTONIC_RAW)
     metrics = Metrics(
-      initializationTimeInMillis: (TimeInterval(timeAfterInit - timeBeforeInit) * 1000)
+      initializationTimeInSeconds: TimeInterval(timeAfterInit - timeBeforeInit)
         / TimeInterval(NSEC_PER_SEC))
 
     super.init()
@@ -256,13 +256,13 @@ extension LlmInference {
   ///
   /// Note: Inherits from `NSObject` for Objective C interoperability.
   @objc(MPPLLMInferenceMetrics) public final class Metrics: NSObject {
-    /// The time it took to initialize the LLM inference engine, in milliseconds.
+    /// The time it took to initialize the LLM inference engine, in seconds.
     /// If you want to include the time it took to load the model weights, set
     /// `LlmInference.Options.waitForWeightUploads` to true.
-    @objc public private(set) var initializationTimeInMillis: TimeInterval
+    @objc public let initializationTimeInSeconds: TimeInterval
 
-    @objc public init(initializationTimeInMillis: TimeInterval) {
-      self.initializationTimeInMillis = initializationTimeInMillis
+    @objc public init(initializationTimeInSeconds: TimeInterval) {
+      self.initializationTimeInSeconds = initializationTimeInSeconds
     }
   }
 }