Merge branch 'main' into bugfix

applications/rag/frontend/container/main.py

@@ -25,6 +25,7 @@
 from langchain.llms import HuggingFaceTextGenInference
 from langchain.prompts import PromptTemplate
 from rai import dlp_filter # Google's Cloud Data Loss Prevention (DLP) API. https://cloud.google.com/security/products/dlp
+from rai import nlp_filter # https://cloud.google.com/natural-language/docs/moderating-text
 from werkzeug.exceptions import HTTPException
 from google.cloud.sql.connector import Connector
 from sentence_transformers import SentenceTransformer
@@ -100,6 +101,11 @@ def getconn() -> pymysql.connections.Connection:
     )
     return pool
 
+@app.route('/get_nlp_status', methods=['GET'])
+def get_nlp_status():
+    nlp_enabled = nlp_filter.is_nlp_api_enabled()
+    return jsonify({"nlpEnabled": nlp_enabled})
+
 @app.route('/get_dlp_status', methods=['GET'])
 def get_dlp_status():
     dlp_enabled = dlp_filter.is_dlp_api_enabled()
@@ -160,7 +166,10 @@ def handlePrompt():
             "context": context,
             "user_prompt": user_prompt
         })
-        if 'inspectTemplate' in data and 'deidentifyTemplate' in data:
+        if 'nlpFilterLevel' in data:
+            if nlp_filter.is_content_inappropriate(response['text'], data.get['nlpFilterLevel']):
+                response['text'] = 'The response is deemed inappropriate for display.'
+        elif 'inspectTemplate' in data and 'deidentifyTemplate' in data:
             inspect_template_path = data['inspectTemplate']
             deidentify_template_path = data['deidentifyTemplate']
             if inspect_template_path != "" and deidentify_template_path != "":

applications/rag/frontend/container/rai/dlp_filter.py

@@ -14,7 +14,7 @@
 
 import os
 import google.cloud.dlp
-from . import retry
+from .retry import retry
 
 # Convert the project id into a full resource id.
 parent = os.environ.get('PROJECT_ID', 'NULL')

applications/rag/frontend/container/rai/nlp_filter.py

@@ -0,0 +1,72 @@
+# Copyright 2024 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.
+
+import os
+import google.cloud.language as language
+from .retry import retry
+
+# Convert the project id into a full resource id.
+parent = os.environ.get('PROJECT_ID', 'NULL')
+
+# Instantiate a nlp client.
+nature_language_client = language.LanguageServiceClient()
+
+def is_nlp_api_enabled():
+  if parent == 'NULL':
+    return False
+  # Check if the DLP API is enabled
+  try:
+    sum_moderation_confidences("test")
+    return True
+  except Exception as e:
+    print(f"Error: {e}")
+    return False
+
+def sum_moderation_confidences(text):
+  document = language.types.Document(
+      content=text, type_=language.types.Document.Type.PLAIN_TEXT
+  )
+
+  request = language.ModerateTextRequest(
+      document=document,
+  )
+  # Detects the sentiment of the text
+  response = nature_language_client.moderate_text(
+      request=request, retry=retry.retry_policy
+  )
+
+  # Parse response and sum the confidences of moderation, the categories are from https://cloud.google.com/natural-language/docs/moderating-text
+  total_confidence = 0.0
+  categories = []
+  for category in response.moderation_categories:
+    total_confidence += category.confidence
+    categories.append(category.name)
+  return total_confidence, categories
+
+def is_content_inappropriate(text, nlp_filter_level):
+  # Define thresholds
+  thresholds = {
+      'low': 1.0,  # 100% confidence for low sensitivity
+      'mid': 0.5,  # 50% confidence for medium sensitivity
+      'high': 0.2,  # 20% confidence for high sensitivity
+  }
+
+  # Map the filter level to a threshold
+  if nlp_filter_level <= 20:
+    threshold = thresholds['high']
+  elif nlp_filter_level <= 50:
+    threshold = thresholds['mid']
+  else:
+    threshold = thresholds['low']
+  return sum_moderation_confidences(text) > threshold

applications/rag/frontend/container/static/script.js

@@ -55,10 +55,12 @@ function onReady() {
         if (filterEnabled) {
             var inspectTemplate = document.getElementById('inspect-template-dropdown').value;
             var deidentifyTemplate = document.getElementById('deidentify-template-dropdown').value;
+            var nlpFilterValue = document.getElementById("nlp-range").value; // Get the NLP filter value
             body = JSON.stringify({
                 prompt: prompt,
                 inspectTemplate: inspectTemplate,
                 deidentifyTemplate: deidentifyTemplate,
+                nlpFilterLevel: nlpFilterValue
             })
         }
 
@@ -107,9 +109,33 @@ function autoResizeTextarea() {
         this.style.height = this.scrollHeight + 'px';
     });
 }
+
+// Function to handle the visibility of filter section
+function toggleNlpFilterSection(nlpEnabled) {
+    var filterOptions = document.getElementById("nlp-filter-section");
+
+    if (nlpEnabled) {
+        filterOptions.style.display = "block";
+    } else {
+        filterOptions.style.display = "none";
+    }
+}
+
+
+function fetchNLPEnabled() {
+    fetch('/get_nlp_status')
+        .then(response => response.json())
+        .then(data => {
+            var nlpEnabled = data.nlpEnabled;
+
+            toggleNlpFilterSection(nlpEnabled);
+        })
+        .catch(error => console.error('Error fetching NLP status:', error))
+}
+
 // Function to handle the visibility of filter section
 function toggleFilterSection(dlpEnabled) {
-    var filterOptions = document.getElementById("filter-section");
+    var filterOptions = document.getElementById("template-section");
 
     if (dlpEnabled) {
         filterOptions.style.display = "block";

applications/rag/frontend/container/static/styles.css

@@ -143,4 +143,13 @@ input[type="submit"]:disabled {
 
 #template-section input[type="checkbox"] {
     margin-top: 10px;
+}
+
+#nlp-filter-section {
+    margin-top: 20px;
+}
+
+#nlp-range {
+    width: 100%;
+    margin: 10px 0;
 }

applications/rag/frontend/container/templates/index.html

@@ -47,6 +47,13 @@
           <!-- Options will be dynamically added here -->
         </select>
       </div>
+
+      <!-- NLP Filter Slider -->
+      <div id="nlp-filter-section">
+        <label for="nlp-range">NLP Filter Level:</label>
+        <input type="range" id="nlp-range" min="0" max="100" value="50">
+      </div>
+
       </div>
     </div>
   </div>

applications/rag/workloads.tfvars

@@ -37,7 +37,6 @@ rag_service_account        = "rag-system-account"
 
 # Creates a google service account & k8s service account & configures workload identity with appropriate permissions.
 # Set to false & update the variable `jupyter_service_account` to use an existing IAM service account.
-create_jupyter_service_account = true
 jupyter_service_account        = "jupyter-system-account"
 
 ## Embeddings table name - change this to the TABLE_NAME used in the notebook.