safety settings in Gemini API

2026-07-21 00:55:21 -04:00 · 2024-07-10 12:14:10 -04:00
parent 1dbdc00f3b
commit dfbec463ac
4 changed files with 110 additions and 1 deletions
@@ -0,0 +1,86 @@
+---
+title: Set Safety Settings When Using Gemini Models
+description: Learn how to set safety settings when using Gemini models in Msty
+---
+
+[Gemini's safety settings](https://ai.google.dev/gemini-api/docs/safety-settings) are designed to control the AI's
+output, ensuring it remains appropriate and safe for various use cases. These settings help filter out potentially
+harmful, offensive, or inappropriate content. The Gemini API offers adjustable safety settings to manage content based
+on predefined categories and sensitive topics. The safety settings can be adjusted to fit different applications, from
+more restrictive settings for general audiences to less restrictive for specialized use cases or adult content.
+
+## Safety Categories
+
+Gemini offers several harm categories, including but not limited to:
+
+* HARM_CATEGORY_UNSPECIFIED
+* HARM_CATEGORY_HATE_SPEECH
+* HARM_CATEGORY_SEXUALLY_EXPLICIT
+* HARM_CATEGORY_HARASSMENT
+* HARM_CATEGORY_DANGEROUS_CONTENT
+
+## Safety Thresholds
+
+For each category, you can set a threshold level:
+
+* BLOCK_NONE: No filtering for this category
+* BLOCK_ONLY_HIGH: Blocks only high-risk content
+* BLOCK_MEDIUM_AND_ABOVE: Blocks medium and high-risk content
+* BLOCK_LOW_AND_ABOVE: Blocks low, medium, and high-risk content
+
+If no safety settings are specified, Gemini uses default settings that aim to provide a balance between openness and
+safety.
+
+## Setting Safety Settings in Msty
+
+Msty doesn't have a UI to set safety settings for Gemini models. However, you can set safety settings using
+the `safety_settings` parameter as a extra model parameters. Here's an example of how you can set safety settings when
+using a Gemini model in Msty:
+
+1. Click on Model Options icon next to model selector and below the chat input.
+2. Open the `Advanced Options` section and locate the `Extra Model Parameters` field at the bottom of the options.
+::image-plain{src="/how-to/gemini-safety-settings/model-options.png" alt="Gemini safety settings in Msty model options"}
+::
+3. In this field, you'll enter a JSON object to configure the safety settings. Use the following structure:
+
+```json
+{
+"safety_settings": 
+    [
+        {
+            "category": "HARM_CATEGORY_NAME",
+            "threshold": "THRESHOLD_LEVEL"
+        }
+    ]
+}
+```
+
+4. Replace "HARM_CATEGORY_NAME" with the specific category you want to adjust.
+5. Replace "THRESHOLD_LEVEL" with the desired threshold level.
+6. You can add multiple categories by including additional objects in the "safety_settings" array.
+
+Example:
+
+```json
+{
+  "safety_settings": [
+    {
+      "category": "HARM_CATEGORY_HARASSMENT",
+      "threshold": "BLOCK_MEDIUM_AND_ABOVE"
+    },
+    {
+      "category": "HARM_CATEGORY_HATE_SPEECH",
+      "threshold": "BLOCK_ONLY_HIGH"
+    },
+    {
+      "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
+      "threshold": "BLOCK_LOW_AND_ABOVE"
+    }
+  ]
+}
+```
+
+8. Apply the changes and run your model with the new safety settings.
+
+Remember, adjusting these settings can impact the type of content Gemini generates, so use them responsibly and in
+accordance with your application's needs and ethical guidelines.