Upload examples

Stay organized with collections Save and categorize content based on your preferences.

After creating an Example Store instance, you can start authoring and uploading examples to it. There's no limit to the number of examples that you can store in an example store instance. Examples become available immediately after you upload them to the Example Store instance.

A few scenarios where you'd need to upload examples include the following:

• The queries are irrelevant to the existing examples.

• The model struggles with some reasoning.

• The available examples don't cover all the functions, outcomes, or reasoning that you expect.

By authoring relevant examples that are in the expected format, you can achieve the following:

• Improve the ability of the LLM to pay attention to the examples and use them, avoiding unexpected changes to response patterns that result from small changes to the prompt.

• Reduce the potential negative impact of adding examples for irrelevant queries.

• The LLM performs as expected to similar queries.

If the LLM shows unexpected behavior or reasoning, you can upload a corrected response to guide the model to follow the expected pattern or reasoning in subsequent requests.

The samples on this page let you author examples based on LLM output. Authoring examples based on the output from an LLM has the following advantages over manually authoring examples:

• Authoring examples based on expected LLM output involves less manual effort.

• By authoring examples based on unexpected LLM behavior, you can directly correct failure cases.

• You can author examples based on responses from well-performing models to improve the behavior of other models. For example, if Gemini 1.5 Pro provides better responses than Gemini 1.5 Flash but with higher latency, you can author examples using those responses to achieve a similar performance at lower latencies using Gemini 1.5 Flash.

Use examples to improve function calling performance

You can use few-shot examples to improve function calling performance by demonstrating the following: * When a particular function is invoked.

• How to extract the arguments to use in your function call.

• How the model responds based on the response returned by the function, or multiple functions in case of multi-step reasoning.

To learn more about function calling, see the Function calling documentation.

Prerequisites

Before you use the Python samples on this page, you must install and initialize the Vertex AI SDK for Python for Example Store in your local Python environment.

1. Run the following command to install the Vertex AI SDK for Python for Example Store.

   pip install --upgrade google-cloud-aiplatform>=1.87.0

2. Use the following code sample to import and initialize the SDK for Example Store.

   import vertexai
   from vertexai.preview import example_stores
   
   vertexai.init(
     project="PROJECT_ID",
     location="LOCATION"
   )
   

   Replace the following:

   • PROJECT_ID: Your project ID.

   • LOCATION: Your region. Only us-central1 is supported.

Upload examples

Use the following samples to upload examples to an Example Store instance. You can upload a maximum of five examples per request.

from vertexai.preview.example_stores import ContentsExample, StoredContentsExample

client = genai.Client(
    http_options=genai_types.HttpOptions(api_version="v1"),
    vertexai=True,
    project="PROJECT_ID",,
    location="LOCATION")

user_content = Content(
    role="user",
    parts=[Part(text="EXAMPLE_QUERY")],
)

response = client.models.generate_content(
    model="MODEL_NAME",
    user_content,
    config=genai_types.GenerateContentConfig(
      tools=[FUNCTION_OR_FUNCTION_DECLARATION]
    )
  )

# Upload example.
example = {
  "contents_example": {
    "contents": [user_content.to_json_dict()],
    "expected_contents": [
      {"content": response.candidates[0].content.to_json_dict()},
      {"content": EXPECTED_FUNCTION_RESPONSE.to_json_dict()},
      {"content": EXPECTED_FINAL_MODEL_RESPONSE.to_json_dict()},
    ],
  },
  "search_key": user_content.parts[0].text,
}
example_store.upsert_examples(examples=[example])

user_content = Content(
    role="user",
    parts=[Part(text="EXAMPLE_QUERY")],
)

example = {
  "contents_example": {
    "contents": [user_content.to_json_dict()],
    "expected_contents": [
      {"content": EXPECTED_FUNCTION_CALL.to_json_dict()},
      {"content": EXPECTED_FUNCTION_RESPONSE.to_json_dict()},
      {"content": EXPECTED_FINAL_MODEL_RESPONSE.to_json_dict()},
    ],
  },
  "search_key": user_content.parts[0].text,
}

example_store.upsert_examples(examples=[example])

POST https://LOCATION-aiplatform.googleapis.com/v1beta1/projects/PROJECT_ID/locations/LOCATION/exampleStores/EXAMPLE_STORE_ID:upsertExamples

{
  "examples": [
      {
          "stored_contents_example": {
              "contents_example": {
                  "contents": [
                      {
                          "role": "user",
                          "parts": [
                              {
                                  "text": "Is there a store in Mountain View, CA that I can visit to try the new Pixel 8 Pro?"
                              }
                          ]
                      }
                  ],
                  "expected_contents": [
                      {
                          "content": {
                              "role": "model",
                              "parts": [
                                  {
                                      "text": ""Yes, there is a store located at 2000 N Shoreline Blvd, Mountain View, CA 94043, US."
                                  }
                              ]
                          }
                      }
                  ]
              },
              "search_key_generation_method": {
                  "last_entry": {}
              }
          }
      }
  ]
}

curl -X POST \
     -H "Authorization: Bearer $(gcloud auth print-access-token)" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d @request.json \
     "https://LOCATION-aiplatform.googleapis.com/v1beta1/projects/PROJECT_ID/locations/LOCATION/exampleStores/EXAMPLE_STORE_ID:upsertExamples"

$cred = gcloud auth print-access-token
$headers = @{ "Authorization" = "Bearer $cred" }

Invoke-WebRequest `
    -Method POST `
    -Headers $headers `
    -ContentType: "application/json; charset=utf-8" `
    -InFile request.json `
    -Uri "https://LOCATION-aiplatform.googleapis.com/v1beta1/projects/PROJECT_ID/locations/LOCATION/exampleStores/EXAMPLE_STORE_ID:upsertExamples" | Select-Object -Expand Content

{
  "results": [
    {
      "example": {
        "exampleId": "exampleTypes/stored_contents_example/examples/EXAMPLE_ID",
        "storedContentsExample": {
          "searchKey": "Is there a store in Mountain View, CA that I can visit to try the new Pixel 8 Pro?",
          "contentsExample": {
            "contents": [
              {
                "role": "user",
                "parts": [
                  {
                    "text": "Is there a store in Mountain View, CA that I can visit to try the new Pixel 8 Pro?"
                  }
                ]
              }
            ],
            "expectedContents": [
              {
                "content": {
                  "role": "model",
                  "parts": [
                    {
                      "text": ""Yes, there is a store located at 2000 N Shoreline Blvd, Mountain View, CA 94043, US."
                    }
                  ]
                }
              }
            ]
          },
          "searchKeyGenerationMethod": {
            "lastEntry": {}
          }
        }
      }
    }
  ]
}

What's next

• Learn how to retrieve examples.