fix: LLM - Make tuning use the global staging bucket if specified

Ark-kun · copybara-github · commit d9ced106b57c · 2023-10-18T16:48:42.000-07:00
PiperOrigin-RevId: 574646855
diff --git a/tests/unit/aiplatform/test_language_models.py b/tests/unit/aiplatform/test_language_models.py
@@ -1723,6 +1723,53 @@ def test_tune_text_generation_model_evaluation_with_only_tensorboard(
             ].runtime_config.parameter_values
             assert pipeline_arguments["tensorboard_resource_id"] == tensorboard_name
 
+    @pytest.mark.parametrize(
+        "job_spec",
+        [_TEST_PIPELINE_SPEC_JSON],
+    )
+    @pytest.mark.parametrize(
+        "mock_request_urlopen",
+        ["https://us-central1-kfp.pkg.dev/proj/repo/pack/latest"],
+        indirect=True,
+    )
+    def test_tune_text_generation_model_staging_bucket(
+        self,
+        mock_pipeline_service_create,
+        mock_pipeline_job_get,
+        mock_pipeline_bucket_exists,
+        job_spec,
+        mock_load_yaml_and_json,
+        mock_gcs_from_string,
+        mock_gcs_upload,
+        mock_request_urlopen,
+        mock_get_tuned_model,
+    ):
+        """Tests that tune_model respects staging_bucket."""
+        TEST_STAGING_BUCKET = "gs://test_staging_bucket/path/"
+        aiplatform.init(staging_bucket=TEST_STAGING_BUCKET)
+
+        with mock.patch.object(
+            target=model_garden_service_client.ModelGardenServiceClient,
+            attribute="get_publisher_model",
+            return_value=gca_publisher_model.PublisherModel(
+                _TEXT_BISON_PUBLISHER_MODEL_DICT
+            ),
+        ):
+            model = language_models.TextGenerationModel.from_pretrained(
+                "text-bison@001"
+            )
+
+            model.tune_model(
+                training_data=_TEST_TEXT_BISON_TRAINING_DF,
+                tuning_job_location="europe-west4",
+                tuned_model_location="us-central1",
+            )
+            call_kwargs = mock_pipeline_service_create.call_args[1]
+            pipeline_arguments = call_kwargs[
+                "pipeline_job"
+            ].runtime_config.parameter_values
+            assert pipeline_arguments["dataset_uri"].startswith(TEST_STAGING_BUCKET)
+
     @pytest.mark.parametrize(
         "job_spec",
         [_TEST_PIPELINE_SPEC_JSON],
diff --git a/vertexai/language_models/_language_models.py b/vertexai/language_models/_language_models.py
@@ -2534,6 +2534,12 @@ def _cancel(self):
 
 
 def _get_tuned_models_dir_uri(model_id: str) -> str:
+    if aiplatform_initializer.global_config.staging_bucket:
+        return (
+            aiplatform_initializer.global_config.staging_bucket
+            + "/tuned_language_models/"
+            + model_id
+        )
     staging_gcs_bucket = (
         gcs_utils.create_gcs_bucket_for_pipeline_artifacts_if_it_does_not_exist()
     )
