Migration guide
AWS Bedrock currently supports Marengo 3.0 and Marengo 2.7. Marengo 2.7 will be deprecated in a future release. This guide helps you migrate from Marengo 2.7 to Marengo 3.0 on AWS Bedrock.
Marengo 3.0 introduces the following key changes to how you invoke the model:
- Nested input structure: Input parameters are now organized inside objects based on the value of the
inputTypeparameter. - Combined text and image inputs: You can now combine text and image in a single request using the new
text_imageinput type. - Parameter updates: Several parameters have changed. For example,
embedding_optionsnow uses different values to specify which types of embeddings to retrieve.
Migration steps
Structural changes
Marengo 3.0 uses a nested structure for all input types. To organize your input parameters, place them inside an object that corresponds to the value of the inputType parameter. For example, if you set inputType to text, you should nest your inputText parameter within an object called text. This structure applies to all input types, including text, image, video, audio, and text_image. The examples below illustrate the differences in structure between Marengo 2.7 and Marengo 3.0 for each input type.
Synchronous processing changes
Text inputs
Marengo 2.7 structure:
Marengo 3.0 structure:
Image inputs
Marengo 2.7 structure:
Marengo 3.0 structure:
Combined text and image inputs
Marengo 2.7 structure: Not supported. You must create separate embddings
Marengo 3.0 structure:
Asynchronous processing changes
Video inputs
Marengo 2.7 structure:
Marengo 3.0 structure:
Audio inputs
Marengo 2.7 structure:
Marengo 3.0 structure:
Image inputs
Marengo 2.7 structure:
Marengo 3.0 structure:
Text inputs
Marengo 2.7 structure:
Marengo 3.0 structure:
Combined text and image inputs
Marengo 2.7 structure: Not supported. You must create separate embddings
Marengo 3.0 structure:
Parameter changes
Marengo 3.0 changes the valid values for the embedding_options parameter. This parameter specifies which types of embeddings to retrieve.
Marengo 2.7 values:
visual-text: Visual embeddings optimized for text searchvisual-image: Visual embeddings optimized for image searchaudio: Embeddings of the audio in the video
Marengo 3.0 values:
visual: Visual embeddings from the videoaudio: Embeddings of the audio in the videotranscription: Embeddings of the transcribed text
Note
Marengo 3.0 introduces additional parameter changes beyond those documented in this guide. For a complete list of request parameters and response fields, see the TwelveLabs Marengo Embed 3.0 page in the AWS Bedrock documentation.