feat: Add API for Gen AI Evaluation in Tuning

Google APIs · copybara-github · commit a6ec08703d66 · 2025-08-12T09:28:57.000-07:00
PiperOrigin-RevId: 794142993
diff --git a/google/cloud/aiplatform/v1beta1/tuning_job.proto b/google/cloud/aiplatform/v1beta1/tuning_job.proto
@@ -20,6 +20,7 @@ import "google/api/field_behavior.proto";
 import "google/api/resource.proto";
 import "google/cloud/aiplatform/v1beta1/content.proto";
 import "google/cloud/aiplatform/v1beta1/encryption_spec.proto";
+import "google/cloud/aiplatform/v1beta1/evaluation_service.proto";
 import "google/cloud/aiplatform/v1beta1/job_state.proto";
 import "google/protobuf/struct.proto";
 import "google/protobuf/timestamp.proto";
@@ -174,6 +175,10 @@ message TuningJob {
   // written to. This field is only available and required for open source
   // models.
   string output_uri = 25 [(google.api.field_behavior) = OPTIONAL];
+
+  // Output only. Evaluation runs for the Tuning Job.
+  repeated EvaluateDatasetRun evaluate_dataset_runs = 32
+      [(google.api.field_behavior) = OUTPUT_ONLY];
 }
 
 // The Model Registry Model and Online Prediction Endpoint associated with
@@ -485,6 +490,10 @@ message SupervisedTuningSpec {
   // checkpoints for SFT. Default is false.
   bool export_last_checkpoint_only = 6 [(google.api.field_behavior) = OPTIONAL];
 
+  // Optional. Evaluation Config for Tuning Job.
+  EvaluationConfig evaluation_config = 5
+      [(google.api.field_behavior) = OPTIONAL];
+
   // Tuning mode.
   TuningMode tuning_mode = 7;
 }
@@ -622,6 +631,32 @@ message VeoTuningSpec {
       [(google.api.field_behavior) = OPTIONAL];
 }
 
+// Evaluation Config for Tuning Job.
+message EvaluationConfig {
+  // Required. The metrics used for evaluation.
+  repeated Metric metrics = 1 [(google.api.field_behavior) = REQUIRED];
+
+  // Required. Config for evaluation output.
+  OutputConfig output_config = 2 [(google.api.field_behavior) = REQUIRED];
+
+  // Optional. Autorater config for evaluation.
+  AutoraterConfig autorater_config = 3 [(google.api.field_behavior) = OPTIONAL];
+}
+
+// Evaluate Dataset Run Result for Tuning Job.
+message EvaluateDatasetRun {
+  // Output only. The operation ID of the evaluation run. Format:
+  // `projects/{project}/locations/{location}/operations/{operation_id}`.
+  string operation_name = 1 [(google.api.field_behavior) = OUTPUT_ONLY];
+
+  // Output only. The checkpoint id used in the evaluation run. Only populated
+  // when evaluating checkpoints.
+  string checkpoint_id = 2 [(google.api.field_behavior) = OUTPUT_ONLY];
+
+  // Output only. The error of the evaluation run if any.
+  google.rpc.Status error = 4 [(google.api.field_behavior) = OUTPUT_ONLY];
+}
+
 // TunedModelCheckpoint for the Tuned Model of a Tuning Job.
 message TunedModelCheckpoint {
   // The ID of the checkpoint.
