Skip to main content

Google

Your connection to any AI Provider or third party app is called a connector. You can find information on setting up google perspective here

modsys supports a wide range of models out of the box. To use a custom AI provider, create a custom module open an issue and we'll put some code together for you.

Notes​

To use the Google Comment Analyzer API, set the PERSPECTIVE_API_KEY environment variable or pass the API key as an argument to the constructor.

Example:

export PERSPECTIVE_API_KEY=your_api_key_here

or

from modsys.client import Modsys

Modsys.use("google_perspective:analyze", google_perspective_api_key="token")

Other Perspective-related environment variables are supported:

  • PERSPECTIVE_CACHE_OUTPUT - You can persist settings for the doNotScore configuration for the API. This setting is responsible for improving the model output over time for the dataset. The default value is True. Users can set the "doNotStore" flag to ensure that all submitted comments are automatically deleted after scores are returned. To set this, you need to override the cache variable in the same way as overriding existing class variables or set the state using PERSPECTIVE_CACHE_OUTPUT.

The Google provider supports the following model formats:

  • google_perspective:analyze - defaults to analyzer
  • google_perspective:suggest - defaults to suggestor
  • google_perspective:<model name> - uses a specific model name (mapped automatically to chat or completion endpoint)

Google Perspective Connector​

The Google Perspective LLM (Language Model) is a tool that enhances LLM output and can be coupled with a Machine Learning validation tool. Users have the option to extend this open-source connector by implementing new attributes to return and more.

Setting up your own API​

To set up your own API, please follow the instructions provided in the Google Perspective API setup guide.

GooglePerspectiveProvider Client​

The GooglePerspectiveProvider is a class that allows you to interact with the Google Perspective API. Here are some important details about this provider:

  • The api_key can be set using the environment variable PERSPECTIVE_API_KEY. It can also be overridden in code by directly setting the api_key_val or more specifically through specifying a secret since it's a class variable.
  • An array of languages to support can be set. The default language is en. Google Perspective API language guide
  • If the API token is not set via the environment variable (default behavior), you will need to specify the token using the secret parameter.
  • There are two model types available: suggest and analyze.
  • You can set your model name during the connection using the use method and specifying google_perspective:<model type>.
  • You can persist settings for the doNotScore configuration for the API. This setting is responsible for improving the model output over time for the dataset. The default value is True. Users can set the "doNotStore" flag to ensure that all submitted comments are automatically deleted after scores are returned. To set this, you need to override the cache variable in the same way as overriding existing class variables or set the state using PERSPECTIVE_CACHE_OUTPUT.
  • To use the Google provider, you need to set the following parameters:
Modsys.use("google_perspective:analyze", google_perspective_api_key=<API KEY(optional)>)

Attributes tracked using Perspective​

The following attributes are tracked using Perspective:

  • TOXICITY: A rude, disrespectful, or unreasonable comment that is likely to make people leave a discussion.
  • SEVERE_TOXICITY: A very hateful, aggressive, disrespectful comment or otherwise very likely to make a user leave a discussion or give up on sharing their perspective. This attribute is much less sensitive to more mild forms of toxicity, such as comments that include positive uses of curse words.
  • INSULT: Insulting, inflammatory, or negative comment towards a person or a group of people.
  • THREAT: Describes an intention to inflict pain, injury, or violence against an individual or group.
  • SEXUALLY_EXPLICIT: Contains references to sexual acts, body parts, or other lewd content.
  • `SPAM: Irrelevant and unsolicited commercial content.

Example Request and Response

Here is an example of the request and response when using the Analyze model: *Suggest model coming soon.

  • Request:
{
  "comment": {
    "text": "string"
  },
  "requestedAttributes": {
    "<ATTR[toxicity, identity attack, threat, sexuality_explicit, flirtation, spam]>": {}
  },
  "languages": ["en"],
  "doNotStore": false,
  "clientToken": "content-id",
  "communityId": "username"
}

Response:

{
  "attributeScores": {
    "string": {
      "summaryScore": {
        "value": 0.5,
        "type": "string"
      },
      "spanScores": [
        {
          "begin": 0,
          "end": 10,
          "score": {
            "value": 0.5,
            "type": "string"
          }
        }
      ]
    }
  },
  "languages": ["string"],
  "clientToken": "string"
}

The response object contains the following information:

  • clientToken (optional): An opaque token that is echoed back in the response.
  • communityId (optional): An opaque identifier associating this comment with a particular community within your platform. If set, this field allows us to differentiate comments from different communities, as each community may have different norms.