Skip to content

Bedrock Titan InvokeModel default values

0

Hello,

Considering the Amazon-Titan request for Bedrock (please see it below), could you please specify which are the default values (when I do no specify any textGenerationConfig) for each AWS model (Premier, Express, Lite), and for the BedrockRuntimeClient.invokeModel java API ? Most interesting are temperature and topP.

{
    "inputText": string,
    "textGenerationConfig": {
        "temperature": float,  
        "topP": float,
        "maxTokenCount": int,
        "stopSequences": [string]
    }
}

Thank you,
Mihai ADAM

asked 2 years ago697 views

1 Answer
1
Accepted Answer

For the Amazon Titan Text models in Amazon Bedrock, the default values for the textGenerationConfig parameters when using the InvokeModel API are as follows:

  1. temperature: The default value is 0.7 for all Titan Text models (Premier, Express, and Lite).

  2. topP: The default value is 0.9 for all Titan Text models.

  3. maxTokenCount: The default value is 512 for all Titan Text models. However, the maximum token limits vary:

  • Titan Text Lite: Maximum of 4,096 tokens
  • Titan Text Express: Maximum of 8,192 tokens
  • Titan Text Premier: Maximum of 3,072 tokens
  1. stopSequences: There is no default value for this parameter. If not specified, the model will continue generating text until it reaches the maxTokenCount or determines the response is complete.

These default values apply when you don't specify any textGenerationConfig in your request, regardless of whether you're using the Java SDK's BedrockRuntimeClient.invokeModel API or any other method to invoke the model.

It's worth noting that while these are the default values, you can adjust them within the allowed ranges to fine-tune the model's output according to your specific needs. The temperature can be set between 0.0 and 1.0, with lower values producing more deterministic outputs. The topP value can also be set between 0.0 and 1.0, with lower values leading to more focused and less diverse responses.
Sources
Amazon Titan Text models - Amazon Bedrock

answered 2 years ago

EXPERT

reviewed 2 years ago

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.