Merge pull request #36 from looker-open-source/uat_llm

Add LLM summaries and Kmeans model objective

Author: ajcrutch

README.md

@@ -25,7 +25,6 @@ The service account used by the BigQuery connection chosen in Step 1 should have
   - BigQuery Job User
   - Vertex AI User
 
-
 #### 3. Create BigQuery Dataset for ML Models
 
 Create a dataset (e.g., `looker_bqml`) in the BigQuery connection's GCP project.
@@ -49,7 +48,7 @@ The application can be installed directly from [Looker Marketplace](https://mark
 
 #### 5. Configure Application with User Attributes
 
-The application uses three [Looker user attributes](https://cloud.google.com/looker/docs/admin-panel-users-user-attributes) to store its configuration settings. The following user attributes are required for the application to work properly. Each user attribute needs to be named exactly as listed below with a data type of `String`. The recommended setting for user access is `None`.
+The application uses three [Looker user attributes](https://cloud.google.com/looker/docs/admin-panel-users-user-attributes) to store its configuration settings. The following user attributes are required for the application to work properly. Each user attribute needs to be named exactly as listed below with a data type of `String`. The recommended setting for user access is `View`.
 
 Create the following user attributes and set their default values.
 
@@ -70,3 +69,34 @@ We recommend creating a new Looker role to easily manage user access to the appl
   - Create a new Looker permission set named `ML Accelerator` containing all the permisions in the [default User permission set](https://cloud.google.com/looker/docs/admin-panel-users-roles#default_permission_sets) AND the `use_sql_runner` permission
   - Create a new Looker role named `ML Accelerator` using the new model and permission set
   - Assign the `ML Accelerator` role to Looker users and groups
+
+#### 7. Setup AI-Generated Model Evaluation Summaries
+
+After release 2.2, the application can use text generating AI to summarize the model evaluation to more clearly communicate model performance. This optional feature requires additional setup.
+
+  ##### 7a: Add an External Connection from Bigquery to Vertex
+  In BigQuery, an [external connection](https://cloud.google.com/bigquery/docs/external-data-sources) is required to connect it to pre-trained models in Vertex AI.  If one is not already set up, you must do so. A tutorial can also be found [here](https://cloud.google.com/bigquery/docs/generate-text-tutorial). 
+1. Under the same gcp project already in use for the application, verify the [BigQuery Connection](https://console.cloud.google.com/apis/library/bigqueryconnection.googleapis.com) and [Vertex AI](https://console.cloud.google.com/apis/library/aiplatform.googleapis.com) APIs are both enabled. 
+2. In BigQuery click “add,” then "Connections to external data sources." 
+3. Select "BigLake and remote function" and use the same location as the dataset already in use by the application
+4. The ID will be the name of your connection. Since it could be used to connect to any number of pre-trained models in vertex it is wise to choose something generic, such as “ext-vertex-ai”
+5. Create the connection
+6. Go to the connection and copy the service account ID. In order to access remote functions from Vertex AI, the [BigQuery connection delegation service agent](https://cloud.google.com/iam/docs/service-agents#bigquery-connection-delegation-service-agent) (of the form bqcx-[#]@gcp-sa-bigquery-condel.iam.gserviceaccount.com) that is associated with this connection must have the "Vertex AI User" role, which can be added in IAM.
+
+  ##### 7b: Create the Remote Text-Generation Model
+
+In BigQuery, enter the following statement in the query editor (this code uses the suggested naming conventions for the the steps above and assumes region is US-Multi). The text-bison@002 model is suggested, but other LLM models with good performance generating text could also be used. The model_name will be later added as a User Attribute value. A suggestion for model_name is "mla-text-bison"
+```
+CREATE OR REPLACE MODEL project_id.dataset_id.model_name
+  REMOTE WITH CONNECTION `us.ext-vertex-ai`
+  OPTIONS (endpoint = 'text-bison@002');
+```
+This will take a few minutes to load and will not return any results. 
+
+  ##### 7c: Update the Relevant User Attribute
+
+  Similar to section 5 above.
+  
+  | **Required User Attribute Name**                                | **Default Value Description**  |
+  |-----------------------------------------------------------------|--------------------------------|
+  | marketplace_bqml_ext_ml_accelerator_generate_text_model_name    | Name chosen in step 7b above   |

bundle.js

@@ -3,13 +3,15 @@ project_name: "marketplace_bqml_ext"
 application: ml-accelerator {
   label: "Machine Learning Accelerator"
   file: "bundle.js"
-  sri_hash: "2Se7ajYLg8GY60c+rlO+X9q3qFnDjV4C7uiYgLWn8iM+/ufBt/+IR+bVOyVc+kmp"
+  sri_hash: "spnMLxVFVZ71aj3VUpreZGBwpLD9+lzINATVhn47QEAgD5WGV85ID3R9wFiSakLq"
   entitlements: {
     core_api_methods: [
       "all_lookml_models",
+      "all_users",
       "create_query",
       "run_query",
       "lookml_model_explore",
+      "get_model",
       "model_fieldname_suggestions",
       "me",
       "user_attribute_user_values",
@@ -25,6 +27,7 @@ application: ml-accelerator {
     scoped_user_attributes: [
       "bigquery_connection_name",
       "bqml_model_dataset_name",
+      "generate_text_model_name",
       "gcp_project",
     ]
   }
@@ -35,10 +38,18 @@ constant: CONNECTION_NAME {
   export: override_required
 }
 
+constant: GCP_PROJECT {
+  value: "{{_user_attributes['marketplace_bqml_ext_ml_accelerator_gcp_project']}}"
+}
+
 constant: BQML_MODEL_DATASET_NAME {
   value: "{{_user_attributes['marketplace_bqml_ext_ml_accelerator_bqml_model_dataset_name']}}"
 }
 
-constant: GCP_PROJECT {
-  value: "{{_user_attributes['marketplace_bqml_ext_ml_accelerator_gcp_project']}}"
+constant: GENERATE_TEXT_MODEL_NAME {
+  value: "{{_user_attributes['marketplace_bqml_ext_ml_accelerator_generate_text_model_name']}}"
 }
+# First create an LLM model in the same dataset as specified in constant "BQML_MODEL_DATASET_NAME", then provide model name here
+# https://cloud.google.com/bigquery/docs/generate-text
+# Also, modify the service account used for the connection to obtain a new permission: bigquery.connections.use
+# This is available to users with role Bigquery Connection User (https://cloud.google.com/iam/docs/understanding-roles#bigquery.connectionUser)

bundle.js.map

@@ -15,33 +15,43 @@
   },
   "user_attributes": {
     "bigquery_connection_name": {
-      "label": "Machine Learning Accelerator Setting: BigQuery Connection Name",
+      "label": "BigQuery Connection Name",
       "description": "The BigQuery connection the application will be allowed to use. Must be the same connection as chosen above.",
       "type": "string",
       "required": true,
       "value_is_hidden": false,
-      "user_can_view": false,
+      "user_can_view": true,
       "user_can_edit": false,
       "default_value": "",
       "value_constraint": "connection"
     },
     "gcp_project": {
-      "label": "Machine Learning Accelerator Setting: GCP Project ID",
+      "label": "GCP Project ID",
       "description": "The GCP project ID for the BigQuery dataset where ML models will be saved.",
       "type": "string",
       "required": true,
       "value_is_hidden": false,
-      "user_can_view": false,
+      "user_can_view": true,
       "user_can_edit": false,
       "default_value": ""
     },
     "bqml_model_dataset_name": {
-      "label": "Machine Learning Accelerator Setting: BQML Model Dataset Name",
+      "label": "BQML Model Dataset Name",
       "description": "The dataset where ML models will be saved. Create a new dataset for BQML models (recommended) or choose the same dataset used for Looker PDTs.",
       "type": "string",
       "required": true,
       "value_is_hidden": false,
-      "user_can_view": false,
+      "user_can_view": true,
+      "user_can_edit": false,
+      "default_value": ""
+    },
+    "generate_text_model_name": {
+      "label": "GenAI Text Model Name",
+      "description": "Name of an LLM model to generate text summaries (optional feature). Must be in same dataset as above. See https://github.com/looker-open-source/app-ml-accelerator/blob/main/README.md#7-setup-ai-generated-evaluation-summaries for setup instructions.",
+      "type": "string",
+      "required": false,
+      "value_is_hidden": false,
+      "user_can_view": true,
       "user_can_edit": false,
       "default_value": ""
     }