// Copyright 2026 Google LLC
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

syntax = "proto3";

package google.cloud.aiplatform.v1;

import "google/api/field_behavior.proto";
import "google/api/resource.proto";
import "google/cloud/aiplatform/v1/content.proto";
import "google/cloud/aiplatform/v1/encryption_spec.proto";
import "google/cloud/aiplatform/v1/evaluation_service.proto";
import "google/cloud/aiplatform/v1/job_state.proto";
import "google/protobuf/timestamp.proto";
import "google/rpc/status.proto";

option csharp_namespace = "Google.Cloud.AIPlatform.V1";
option go_package = "cloud.google.com/go/aiplatform/apiv1/aiplatformpb;aiplatformpb";
option java_multiple_files = true;
option java_outer_classname = "TuningJobProto";
option java_package = "com.google.cloud.aiplatform.v1";
option php_namespace = "Google\\Cloud\\AIPlatform\\V1";
option ruby_package = "Google::Cloud::AIPlatform::V1";

// Represents a TuningJob that runs with Google owned models.
message TuningJob {
  option (google.api.resource) = {
    type: "aiplatform.googleapis.com/TuningJob"
    pattern: "projects/{project}/locations/{location}/tuningJobs/{tuning_job}"
    plural: "tuningJobs"
    singular: "tuningJob"
  };

  oneof source_model {
    // The base model that is being tuned. See [Supported
    // models](https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/tuning#supported_models).
    string base_model = 4;

    // The pre-tuned model for continuous tuning.
    PreTunedModel pre_tuned_model = 31;
  }

  oneof tuning_spec {
    // Tuning Spec for Supervised Fine Tuning.
    SupervisedTuningSpec supervised_tuning_spec = 5;
  }

  // Output only. Identifier. Resource name of a TuningJob. Format:
  // `projects/{project}/locations/{location}/tuningJobs/{tuning_job}`
  string name = 1 [
    (google.api.field_behavior) = IDENTIFIER,
    (google.api.field_behavior) = OUTPUT_ONLY
  ];

  // Optional. The display name of the
  // [TunedModel][google.cloud.aiplatform.v1.Model]. The name can be up to 128
  // characters long and can consist of any UTF-8 characters. For continuous
  // tuning, tuned_model_display_name will by default use the same display name
  // as the pre-tuned model. If a new display name is provided, the tuning job
  // will create a new model instead of a new version.
  string tuned_model_display_name = 2 [(google.api.field_behavior) = OPTIONAL];

  // Optional. The description of the
  // [TuningJob][google.cloud.aiplatform.v1.TuningJob].
  string description = 3 [(google.api.field_behavior) = OPTIONAL];

  // Output only. The detailed state of the job.
  JobState state = 6 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Time when the
  // [TuningJob][google.cloud.aiplatform.v1.TuningJob] was created.
  google.protobuf.Timestamp create_time = 7
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Time when the
  // [TuningJob][google.cloud.aiplatform.v1.TuningJob] for the first time
  // entered the `JOB_STATE_RUNNING` state.
  google.protobuf.Timestamp start_time = 8
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Time when the TuningJob entered any of the following
  // [JobStates][google.cloud.aiplatform.v1.JobState]: `JOB_STATE_SUCCEEDED`,
  // `JOB_STATE_FAILED`, `JOB_STATE_CANCELLED`, `JOB_STATE_EXPIRED`.
  google.protobuf.Timestamp end_time = 9
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Time when the
  // [TuningJob][google.cloud.aiplatform.v1.TuningJob] was most recently
  // updated.
  google.protobuf.Timestamp update_time = 10
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Only populated when job's state is `JOB_STATE_FAILED` or
  // `JOB_STATE_CANCELLED`.
  google.rpc.Status error = 11 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Optional. The labels with user-defined metadata to organize
  // [TuningJob][google.cloud.aiplatform.v1.TuningJob] and generated resources
  // such as [Model][google.cloud.aiplatform.v1.Model] and
  // [Endpoint][google.cloud.aiplatform.v1.Endpoint].
  //
  // Label keys and values can be no longer than 64 characters
  // (Unicode codepoints), can only contain lowercase letters, numeric
  // characters, underscores and dashes. International characters are allowed.
  //
  // See https://goo.gl/xmQnxf for more information and examples of labels.
  map<string, string> labels = 12 [(google.api.field_behavior) = OPTIONAL];

  // Output only. The Experiment associated with this
  // [TuningJob][google.cloud.aiplatform.v1.TuningJob].
  string experiment = 13 [
    (google.api.field_behavior) = OUTPUT_ONLY,
    (google.api.resource_reference) = {
      type: "aiplatform.googleapis.com/Context"
    }
  ];

  // Output only. The tuned model resources associated with this
  // [TuningJob][google.cloud.aiplatform.v1.TuningJob].
  TunedModel tuned_model = 14 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. The tuning data statistics associated with this
  // [TuningJob][google.cloud.aiplatform.v1.TuningJob].
  TuningDataStats tuning_data_stats = 15
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Customer-managed encryption key options for a TuningJob. If this is set,
  // then all resources created by the TuningJob will be encrypted with the
  // provided encryption key.
  EncryptionSpec encryption_spec = 16;

  // The service account that the tuningJob workload runs as.
  // If not specified, the Vertex AI Secure Fine-Tuned Service Agent in the
  // project will be used. See
  // https://cloud.google.com/iam/docs/service-agents#vertex-ai-secure-fine-tuning-service-agent
  //
  // Users starting the pipeline must have the `iam.serviceAccounts.actAs`
  // permission on this service account.
  string service_account = 22;

  // 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
// this [TuningJob][google.cloud.aiplatform.v1.TuningJob].
message TunedModel {
  // Output only. The resource name of the TunedModel. Format:
  //
  // `projects/{project}/locations/{location}/models/{model}@{version_id}`
  //
  // When tuning from a base model, the version ID will be 1.
  //
  // For continuous tuning, if the provided tuned_model_display_name is set and
  // different from parent model's display name, the tuned model will have a new
  // parent model with version 1. Otherwise the version id will be incremented
  // by 1 from the last version ID in the parent model. E.g.,
  //
  // `projects/{project}/locations/{location}/models/{model}@{last_version_id +
  // 1}`
  string model = 1 [
    (google.api.field_behavior) = OUTPUT_ONLY,
    (google.api.resource_reference) = {
      type: "aiplatform.googleapis.com/Model"
    }
  ];

  // Output only. A resource name of an Endpoint. Format:
  // `projects/{project}/locations/{location}/endpoints/{endpoint}`.
  string endpoint = 2 [
    (google.api.field_behavior) = OUTPUT_ONLY,
    (google.api.resource_reference) = {
      type: "aiplatform.googleapis.com/Endpoint"
    }
  ];

  // Output only. The checkpoints associated with this TunedModel.
  // This field is only populated for tuning jobs that enable intermediate
  // checkpoints.
  repeated TunedModelCheckpoint checkpoints = 3
      [(google.api.field_behavior) = OUTPUT_ONLY];
}

// Dataset distribution for Supervised Tuning.
message SupervisedTuningDatasetDistribution {
  // Dataset bucket used to create a histogram for the distribution given a
  // population of values.
  message DatasetBucket {
    // Output only. Number of values in the bucket.
    double count = 1 [(google.api.field_behavior) = OUTPUT_ONLY];

    // Output only. Left bound of the bucket.
    double left = 2 [(google.api.field_behavior) = OUTPUT_ONLY];

    // Output only. Right bound of the bucket.
    double right = 3 [(google.api.field_behavior) = OUTPUT_ONLY];
  }

  // Output only. Sum of a given population of values.
  int64 sum = 1 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Sum of a given population of values that are billable.
  int64 billable_sum = 9 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. The minimum of the population values.
  double min = 2 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. The maximum of the population values.
  double max = 3 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. The arithmetic mean of the values in the population.
  double mean = 4 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. The median of the values in the population.
  double median = 5 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. The 5th percentile of the values in the population.
  double p5 = 6 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. The 95th percentile of the values in the population.
  double p95 = 7 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Defines the histogram bucket.
  repeated DatasetBucket buckets = 8
      [(google.api.field_behavior) = OUTPUT_ONLY];
}

// Tuning data statistics for Supervised Tuning.
message SupervisedTuningDataStats {
  // Output only. Number of examples in the tuning dataset.
  int64 tuning_dataset_example_count = 1
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Number of tuning characters in the tuning dataset.
  int64 total_tuning_character_count = 2
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Number of billable characters in the tuning dataset.
  int64 total_billable_character_count = 3
      [deprecated = true, (google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Number of billable tokens in the tuning dataset.
  int64 total_billable_token_count = 9
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Number of tuning steps for this Tuning Job.
  int64 tuning_step_count = 4 [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Dataset distributions for the user input tokens.
  SupervisedTuningDatasetDistribution user_input_token_distribution = 5
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Dataset distributions for the user output tokens.
  SupervisedTuningDatasetDistribution user_output_token_distribution = 6
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Dataset distributions for the messages per example.
  SupervisedTuningDatasetDistribution user_message_per_example_distribution = 7
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. Sample user messages in the training dataset uri.
  repeated Content user_dataset_examples = 8
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. The number of examples in the dataset that have been dropped.
  // An example can be dropped for reasons including: too many tokens, contains
  // an invalid image, contains too many images, etc.
  int64 total_truncated_example_count = 10
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. A partial sample of the indices (starting from 1) of the
  // dropped examples.
  repeated int64 truncated_example_indices = 11
      [(google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. For each index in `truncated_example_indices`, the user-facing
  // reason why the example was dropped.
  repeated string dropped_example_reasons = 12
      [(google.api.field_behavior) = OUTPUT_ONLY];
}

// The tuning data statistic values for
// [TuningJob][google.cloud.aiplatform.v1.TuningJob].
message TuningDataStats {
  oneof tuning_data_stats {
    // The SFT Tuning data stats.
    SupervisedTuningDataStats supervised_tuning_data_stats = 1;
  }
}

// Hyperparameters for SFT.
message SupervisedHyperParameters {
  // Supported adapter sizes for tuning.
  enum AdapterSize {
    // Adapter size is unspecified.
    ADAPTER_SIZE_UNSPECIFIED = 0;

    // Adapter size 1.
    ADAPTER_SIZE_ONE = 1;

    // Adapter size 2.
    ADAPTER_SIZE_TWO = 6;

    // Adapter size 4.
    ADAPTER_SIZE_FOUR = 2;

    // Adapter size 8.
    ADAPTER_SIZE_EIGHT = 3;

    // Adapter size 16.
    ADAPTER_SIZE_SIXTEEN = 4;

    // Adapter size 32.
    ADAPTER_SIZE_THIRTY_TWO = 5;
  }

  // Optional. Number of complete passes the model makes over the entire
  // training dataset during training.
  int64 epoch_count = 1 [(google.api.field_behavior) = OPTIONAL];

  // Optional. Multiplier for adjusting the default learning rate.
  double learning_rate_multiplier = 2 [(google.api.field_behavior) = OPTIONAL];

  // Optional. Adapter size for tuning.
  AdapterSize adapter_size = 3 [(google.api.field_behavior) = OPTIONAL];
}

// Tuning Spec for Supervised Tuning for first party models.
message SupervisedTuningSpec {
  // Required. Cloud Storage path to file containing training dataset for
  // tuning. The dataset must be formatted as a JSONL file.
  string training_dataset_uri = 1 [(google.api.field_behavior) = REQUIRED];

  // Optional. Cloud Storage path to file containing validation dataset for
  // tuning. The dataset must be formatted as a JSONL file.
  string validation_dataset_uri = 2 [(google.api.field_behavior) = OPTIONAL];

  // Optional. Hyperparameters for SFT.
  SupervisedHyperParameters hyper_parameters = 3
      [(google.api.field_behavior) = OPTIONAL];

  // Optional. If set to true, disable intermediate checkpoints for SFT and only
  // the last checkpoint will be exported. Otherwise, enable intermediate
  // 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];
}

// TunedModel Reference for legacy model migration.
message TunedModelRef {
  // The Tuned Model Reference for the model.
  oneof tuned_model_ref {
    // Support migration from model registry.
    string tuned_model = 1 [(google.api.resource_reference) = {
      type: "aiplatform.googleapis.com/Model"
    }];

    // Support migration from tuning job list page, from gemini-1.0-pro-002
    // to 1.5 and above.
    string tuning_job = 2 [(google.api.resource_reference) = {
      type: "aiplatform.googleapis.com/TuningJob"
    }];

    // Support migration from tuning job list page, from bison model to gemini
    // model.
    string pipeline_job = 3 [(google.api.resource_reference) = {
      type: "aiplatform.googleapis.com/PipelineJob"
    }];
  }
}

// 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];

  // Optional. Configuration options for inference generation and outputs.
  // If not set, default generation parameters are used.
  GenerationConfig inference_generation_config = 5
      [(google.api.field_behavior) = OPTIONAL];
}

// Evaluate Dataset Run Result for Tuning Job.
message EvaluateDatasetRun {
  // Output only. Deprecated: The updated architecture uses evaluation_run
  // instead.
  string operation_name = 1
      [deprecated = true, (google.api.field_behavior) = OUTPUT_ONLY];

  // Output only. The resource name of the evaluation run. Format:
  // `projects/{project}/locations/{location}/evaluationRuns/{evaluation_run_id}`.
  string evaluation_run = 5 [(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. Results for EvaluationService.
  EvaluateDatasetResponse evaluate_dataset_response = 3
      [(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.
  string checkpoint_id = 1;

  // The epoch of the checkpoint.
  int64 epoch = 2;

  // The step of the checkpoint.
  int64 step = 3;

  // The Endpoint resource name that the checkpoint is deployed to. Format:
  // `projects/{project}/locations/{location}/endpoints/{endpoint}`.
  string endpoint = 4;
}

// A pre-tuned model for continuous tuning.
message PreTunedModel {
  // The resource name of the Model.
  // E.g., a model resource name with a specified version id or alias:
  //
  // `projects/{project}/locations/{location}/models/{model}@{version_id}`
  //
  // `projects/{project}/locations/{location}/models/{model}@{alias}`
  //
  // Or, omit the version id to use the default version:
  //
  // `projects/{project}/locations/{location}/models/{model}`
  string tuned_model_name = 1 [(google.api.resource_reference) = {
    type: "aiplatform.googleapis.com/Model"
  }];

  // Optional. The source checkpoint id. If not specified, the default
  // checkpoint will be used.
  string checkpoint_id = 2 [(google.api.field_behavior) = OPTIONAL];

  // Output only. The name of the base model this
  // [PreTunedModel][google.cloud.aiplatform.v1.PreTunedModel] was tuned from.
  string base_model = 3 [(google.api.field_behavior) = OUTPUT_ONLY];
}