Generation
AI generation routes are a request-response API used to generate structured output from AI models. Examples of generation routes include:
- generated structured data from unstructured input
- summarization
Under the hood, a generation route is an AWS AppSync query that ensures the AI model responds with the response type defined for the route.
Generate Typed Objects
Schema Definition
const schema = a.schema({ generateRecipe: a.generation({ aiModel: a.ai.model('Claude 3 Haiku'), systemPrompt: 'You are a helpful assistant that generates recipes.', }) .arguments({ description: a.string() }) .returns( a.customType({ name: a.string(), ingredients: a.string().array(), instructions: a.string(), }) ) .authorization((allow) => allow.authenticated())});Data Client Request
const description = 'I would like to bake a birthday cake for my friend. She has celiac disease and loves chocolate.'const { data, errors } = await client.generations .generateRecipe({ description })
/**Example response:{ "name": "Gluten-Free Chocolate Birthday Cake", "ingredients": [ "gluten-free all-purpose flour", "cocoa powder", "granulated sugar", "baking powder", "baking soda", "salt", "eggs", "milk", "vegetable oil", "vanilla extract" ], "instructions": "1. Preheat oven to 350°F. Grease and flour two 9-inch round baking pans.\n2. In a medium bowl, whisk together the gluten-free flour, cocoa powder, sugar, baking powder, baking soda and salt.\n3. In a separate bowl, beat the eggs. Then add the milk, oil and vanilla and mix well.\n4. Gradually add the wet ingredients to the dry ingredients and mix until just combined. Do not over mix.\n5. Divide the batter evenly between the prepared pans.\n6. Bake for 30-35 minutes, until a toothpick inserted in the center comes out clean.\n7. Allow cakes to cool in pans for 10 minutes, then transfer to a wire rack to cool completely.\n8. Frost with your favorite gluten-free chocolate frosting."}*/Generate Scalar Types
Schema Definition
const schema = ({ summarize: a.generation({ aiModel: a.ai.model('Claude 3 Haiku'), systemPrompt: 'Provide an accurate, clear, and concise summary of the input provided' }) .arguments({ input: a.string() }) .returns(a.string()) .authorization((allow) => allow.guest()),});Data Client Request
const { data: summary, errors } = await client.generations .summarize({ input })Setting Inference Parameters
You can influence response generation by setting inference parameters for the AI model. This ability to control the randomness and diversity of responses is useful for generating responses that are tailored to your needs.
More information about inference parameters.
Inference Parameters
const schema = a.schema({ generateHaiku: a.generation({ aiModel: a.ai.model('Claude 3 Haiku'), systemPrompt: 'You are a helpful assistant that generates haikus.', inferenceConfiguration: { maxTokens: 1000, temperature: 0.5, topP: 0.9, } }),});Limitations
1. Generation routes do not support referencing models
For example, the following schema defines a Recipe model, but this model cannot be used as the return type of a generation route.
Invalid Model Reference
const schema = a.schema({ Recipe: a.model({ name: a.string(), ingredients: a.string().array(), instructions: a.string(), }), generateRecipe: a.generation({ aiModel: a.ai.model('Claude 3 Haiku'), systemPrompt: 'You are a helpful assistant that generates recipes.', }) .arguments({ description: a.string() }) .returns(a.ref('Recipe')) // ❌ Invalid .authorization((allow) => allow.authenticated()),});You can, however, reference custom types. Here's an example of a custom type that can be used as the return type of a generation route.
Valid Custom Type Reference
const schema = a.schema({ Recipe: a.customType({ name: a.string(), ingredients: a.string().array(), instructions: a.string(), }), generateRecipe: a.generation({ aiModel: a.ai.model('Claude 3 Haiku'), systemPrompt: 'You are a helpful assistant that generates recipes.', }) .arguments({ description: a.string() }) .returns(a.ref('Recipe')) // ✅ Valid .authorization((allow) => allow.authenticated()),});2. Generation routes do not support some required types.
The following AppSync scalar types are not supported as required fields in response types:
AWSEmailAWSDateAWSTimeAWSDateTimeAWSTimestampAWSPhoneAWSURLAWSIPAddress
Unsupported Required Type
const schema = a.schema({ generateUser: a.generation({ aiModel: a.ai.model('Claude 3 Haiku'), systemPrompt: 'You are a helpful assistant that generates users.', }) .arguments({ description: a.string() }) .returns( a.customType({ name: a.string(), email: a.email().required(), // ❌ Required field with unsupported type dateOfBirth: a.date().required(), // ❌ Required field with unsupported type }) ) .authorization((allow) => allow.authenticated()),});