AWS Amazon Titan

Amazon Titan

When it comes to Amazon Titan, the format of the prompt depends on the intended purpose. As you will recall from a previous lesson, the basic elements of a prompt are as follows:

• Instruction: Describes what you are asking of the model, such as summarization or question answering.
• Context: Sets the scene for the model, such as making suggestions like this one: "The following data is a medical text."
• Input: Provides the actual input text for summaries or retrieval tasks.
• Output indicator: Constrains the output from the model. For example, you may want the model to limit the output to a certain number of words or sentences while summarizing a passage.

Review the following best practices for using Amazon Titan.

Specify output length

Be specific about the number of sentences, bullet points, and paragraphs you want the model to generate in the output. Instead of asking for a response with a certain number of words, instruct the model to return, for example, "a summary in exactly three sentences" or "a list with four bullet points." Clearly stating these format requirements will help achieve more consistent and controlled results from Amazon Titan and other language models.

Provide simple, clear, and complete instructions

Amazon Titan generates the best responses when you include an unambiguous expectation of the task in your prompt.

Add context to the prompt for more detailed answers

You achieve better results when you add context and provide the instruction or output indicator at the end of the prompt. The following example demonstrates this approach.

Provide default output when necessary

You can provide default output for the model to present if it's unsure about how to answer.

This prevents the model from returning answers that sound correct but might not answer the question.

The following is an example.

Use separator characters for API calls

For Amazon Titan models, adding \n (new line character) at the end of a prompt will likely improve the model's performance. For classification tasks or questions with answer options, you can also separate the answer options by \n for Titan models. The following example shows how newline characters between choices and at the end of a prompt help you produce the desired response with Amazon Titan.

Personalize responses

You can personalize responses by including a role in your prompt. This technique helps you generate personalized responses. The following example demonstrates this approach.

Generate code

You can generate code samples, such as SQL, Python, HTML, or Java. The following example shows this capability.

# Python
import boto3
# Creates a boto3 client object
s3 = boto3.client('s3')
def download_file(bucket_name, object_key, file_path):
    try:
        # Downloads the S3 file to the specified file path
        s3.download_file(bucket_name, object_key, file_path)
    except Exception as ex:
        print(ex)

# Specifies the values of bucket_name, object_key, and file_path
bucket_name = "samplebucketname"
object_key = "sampleobjectkey"
file_path = "/samplefilepath"
download_file(bucket_name, object_key, file_path)