Add EmbeddingValue union type and Base64 support for embeddings #519

yoshioterada · 2025-06-30T12:51:26Z

This implementation adds support for Base64-encoded embeddings as the default response format, while maintaining complete backward compatibility with existing List<Float> usage.

1. Default Behavior Change

New Default: Embedding requests now default to Base64 encoding format
Backward Compatibility: Existing code using embedding() method continues to work unchanged
Performance: Base64 encoding reduces network payload size significantly

Introduces the EmbeddingValue class to support both float list and base64-encoded embedding data, enabling efficient handling and backward compatibility. Embedding, EmbeddingCreateParams, and related classes are updated to use EmbeddingValue, with automatic decoding and encoding between formats. Adds EmbeddingDefaults for global default encoding configuration, and comprehensive tests for new behaviors and compatibility.

Improvement from previous PR

Overload: Due to the spec of the Kotlin, it was not possible to implement with overload method.
Immutable: Modified to Immutable class
Private Constructor: Created the class with private constructor

This PR is the fix of the issue #211
This issue is related to the PR of #303

This implementation adds support for Base64-encoded embeddings as the default response format, while maintaining complete backward compatibility with existing `List<Float>` usage. ### 1. Default Behavior Change - **New Default**: Embedding requests now default to Base64 encoding format - **Backward Compatibility**: Existing code using `embedding()` method continues to work unchanged - **Performance**: Base64 encoding reduces network payload size significantly Introduces the EmbeddingValue class to support both float list and base64-encoded embedding data, enabling efficient handling and backward compatibility. Embedding, EmbeddingCreateParams, and related classes are updated to use EmbeddingValue, with automatic decoding and encoding between formats. Adds EmbeddingDefaults for global default encoding configuration, and comprehensive tests for new behaviors and compatibility.

yoshioterada · 2025-06-30T13:34:45Z

When verifying the operation in the Azure OpenAI environment, I was able to confirm the operation with the following code:

package com.openai.example;

import com.openai.azure.AzureOpenAIServiceVersion;
import com.openai.azure.credential.AzureApiKeyCredential;
import com.openai.client.OpenAIClient;
import com.openai.client.OpenAIClientAsync;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.client.okhttp.OpenAIOkHttpClientAsync;
import com.openai.models.embeddings.CreateEmbeddingResponse;
import com.openai.models.embeddings.EmbeddingCreateParams;
import com.openai.models.embeddings.EmbeddingModel;
import com.openai.models.embeddings.EmbeddingValue;
import com.openai.services.blocking.EmbeddingService;
import java.util.ArrayList;
import java.util.List;
import java.util.concurrent.CompletableFuture;
import java.util.concurrent.CountDownLatch;

/**
 * Sample code using Azure OpenAI Embedding API
 * Demonstrates how to retrieve and process embedding data in Base64 and Float formats
 */
public final class EmbeddingsExampleAzure {
    private EmbeddingsExampleAzure() {}

    // Azure OpenAI endpoint and key configuration
    // Please replace with actual values
    private static final String AZURE_OPENAI_ENDPOINT = "https://***********.openai.azure.com";
    private static final String AZURE_OPENAI_KEY = "***********";

    private static OpenAIClient client;

    public static void main(String[] args) {
        // Initialize Azure OpenAI client
        client = OpenAIOkHttpClient.builder()
                .baseUrl(AZURE_OPENAI_ENDPOINT)
                .credential(AzureApiKeyCredential.create(AZURE_OPENAI_KEY))
                .azureServiceVersion(AzureOpenAIServiceVersion.getV2024_02_15_PREVIEW())
                .build();

        EmbeddingsExampleAzure example = new EmbeddingsExampleAzure();
        example.basicSample();
        example.multipleDataSample();
        example.asyncSample();
    }

    /**
     * Basic embedding retrieval sample
     * Demonstrates usage of default format, Float format, and Base64 format
     */
    public void basicSample() {
        EmbeddingService embeddings = client.embeddings();
        String singlePoem = "In the quiet night, stars whisper secrets, dreams take flight.";

        System.out.println("=== Basic Embedding Sample ===");

        // 1. Default format (Base64 is default)
        System.out.println("\n1. Getting embeddings in default format:");
        EmbeddingCreateParams embeddingCreateParams = EmbeddingCreateParams.builder()
                .input(singlePoem)
                .model(EmbeddingModel.TEXT_EMBEDDING_3_SMALL.asString())
                .build();

        embeddings.create(embeddingCreateParams).data().forEach(embedding -> {
            System.out.println("Embedding (default format): " + embedding.toString());

            // Use EmbeddingValue to check the original format
            EmbeddingValue embeddingValue = embedding.embeddingValue();
            if (embeddingValue.isBase64String()) {
                System.out.println(
                        "Received in Base64 format: " + embeddingValue.base64String().substring(0, 50) + "...");
            } else if (embeddingValue.isFloatList()) {
                System.out.println("Received in Float format: " + embeddingValue.floatList().size() + " elements");
            }

            // embedding() method always returns List<Float> (Base64 is automatically decoded)
            List<Float> floats = embedding.embedding();
            System.out.println(
                    "Retrieved as Float array: " + floats.size() + " elements, first 5: " + floats.subList(0, Math.min(5, floats.size())));
        });

        System.out.println("\n------------------------------------------------");

        // 2. Explicitly specify Float format
        System.out.println("\n2. Explicitly specifying Float format:");
        EmbeddingCreateParams embeddingCreateParams2 = EmbeddingCreateParams.builder()
                .input(singlePoem)
                .model(EmbeddingModel.TEXT_EMBEDDING_3_SMALL.asString())
                .encodingFormat(EmbeddingCreateParams.EncodingFormat.FLOAT)
                .build();

        embeddings.create(embeddingCreateParams2).data().forEach(embedding -> {
            EmbeddingValue embeddingValue = embedding.embeddingValue();
            if (embeddingValue.isFloatList()) {
                System.out.println("Received in Float format: " + embeddingValue.floatList().size() + " elements");
                System.out.println("First 5 values: "
                        + embeddingValue
                                .floatList()
                                .subList(
                                        0,
                                        Math.min(5, embeddingValue.floatList().size())));
            }

            // Can also convert to Base64 format
            String base64 = embeddingValue.asBase64String();
            System.out.println("Converted to Base64 format: " + base64.substring(0, 50) + "...");
        });

        System.out.println("\n------------------------------------------------");

        // 3. Explicitly specify Base64 format
        System.out.println("\n3. Explicitly specifying Base64 format:");
        EmbeddingCreateParams embeddingCreateParams3 = EmbeddingCreateParams.builder()
                .input(singlePoem)
                .model(EmbeddingModel.TEXT_EMBEDDING_3_SMALL.asString())
                .encodingFormat(EmbeddingCreateParams.EncodingFormat.BASE64)
                .build();

        embeddings.create(embeddingCreateParams3).data().forEach(embedding -> {
            EmbeddingValue embeddingValue = embedding.embeddingValue();
            if (embeddingValue.isBase64String()) {
                System.out.println(
                        "Received in Base64 format: " + embeddingValue.base64String().substring(0, 50) + "...");
            }

            // Automatically convert to Float array
            List<Float> floats = embeddingValue.asFloatList();
            System.out.println(
                    "Auto-converted to Float array: " + floats.size() + " elements, first 5: " + floats.subList(0, Math.min(5, floats.size())));
        });

        System.out.println("\n================================================");
    }

    /**
     * Multiple data embedding retrieval sample
     */
    public void multipleDataSample() {
        EmbeddingService embeddings = client.embeddings();

        System.out.println("\n=== Multiple Data Embedding Sample ===");

        getPoems().forEach(poem -> {
            System.out.println("\nPoem (start): " + poem);

            EmbeddingCreateParams embeddingCreateParams = EmbeddingCreateParams.builder()
                    .input(poem)
                    .model(EmbeddingModel.TEXT_EMBEDDING_3_SMALL.asString())
                    .build(); // Use default format (Base64)

            embeddings.create(embeddingCreateParams).data().forEach(embedding -> {
                List<Float> floats = embedding.embedding();
                System.out.println("Embedding (default): " + floats.size() + " dimensions");

                // Check original format
                EmbeddingValue embeddingValue = embedding.embeddingValue();
                if (embeddingValue.isBase64String()) {
                    System.out.println("Original format: Base64");
                } else {
                    System.out.println("Original format: Float array");
                }
            });
            System.out.println("Poem (end)");
        });

        System.out.println("\n================================================");
    }

    /**
     * Asynchronous embedding retrieval sample
     */
    public void asyncSample() {
        System.out.println("\n=== Asynchronous Embedding Sample ===");

        CountDownLatch latch = new CountDownLatch(1);
        try {
            OpenAIClientAsync asyncClient = OpenAIOkHttpClientAsync.builder()
                    .baseUrl(AZURE_OPENAI_ENDPOINT)
                    .credential(AzureApiKeyCredential.create(AZURE_OPENAI_KEY))
                    .azureServiceVersion(AzureOpenAIServiceVersion.getV2024_02_15_PREVIEW())
                    .build();

            CompletableFuture<CreateEmbeddingResponse> completableFuture = asyncClient
                    .embeddings()
                    .create(EmbeddingCreateParams.builder()
                            .input("The quick brown fox jumped over the lazy dog")
                            .model(EmbeddingModel.TEXT_EMBEDDING_3_SMALL)
                            .encodingFormat(EmbeddingCreateParams.EncodingFormat.FLOAT)
                            .user("user-1234")
                            .build());

            completableFuture
                    .thenAccept(response -> {
                        response.validate();
                        response.data().forEach(embedding -> {
                            System.out.println("Asynchronous embedding retrieval completed:");
                            System.out.println("Embedding info: " + embedding.toString());

                            EmbeddingValue embeddingValue = embedding.embeddingValue();
                            if (embeddingValue.isFloatList()) {
                                System.out.println(
                                        "Float format: " + embeddingValue.floatList().size() + " dimensions");
                            }

                            // Visitor pattern usage example
                            String result = embeddingValue.accept(new EmbeddingValue.Visitor<String>() {
                                @Override
                                public String visitFloatList(List<Float> floatList) {
                                    return "Processing Float array: " + floatList.size() + " elements";
                                }

                                @Override
                                public String visitBase64String(String base64String) {
                                    return "Processing Base64 string: " + base64String.length() + " characters";
                                }
                            });
                            System.out.println("Visitor pattern result: " + result);

                            latch.countDown();
                        });
                    })
                    .exceptionally(ex -> {
                        System.err.println("Error: " + ex.getMessage());
                        latch.countDown();
                        return null;
                    });

            latch.await();
            System.out.println("Asynchronous processing completed");

        } catch (InterruptedException e) {
            Thread.currentThread().interrupt();
            System.err.println("Processing was interrupted");
        }

        System.out.println("\n================================================");
        System.out.println("All samples execution completed");
    }

    /**
     * Get sample poems list
     */
    private List<String> getPoems() {
        List<String> poems = new ArrayList<>();
        poems.add("In the quiet night, stars whisper secrets, dreams take flight.");
        poems.add("Beneath the moon's glow, shadows dance, hearts begin to know.");
        poems.add("Waves crash on the shore, time stands still, love forevermore.");
        poems.add("Autumn leaves fall, painting the ground, nature's final call.");
        poems.add("Morning dew glistens, a new day dawns, hope always listens.");
        poems.add("Mountains stand tall, silent guardians, witnessing it all.");
        poems.add("In a field of green, flowers bloom bright, a serene scene.");
        poems.add("Winter's chill bites, fireside warmth, cozy, long nights.");
        poems.add("Spring's gentle breeze, life awakens, hearts find ease.");
        poems.add("Sunset hues blend, day meets night, a perfect end.");
        return poems;
    }
}

Deleted invalid EmbeddingDefaults.kt

TomerAberbach · 2025-07-01T16:40:59Z

openai-java-core/src/main/kotlin/com/openai/models/embeddings/Embedding.kt

@@ -22,6 +22,7 @@ import kotlin.jvm.optionals.getOrNull
 class Embedding
 private constructor(
    private val embedding: JsonField<List<Float>>,


Since the new EmbeddingValue class supports List<Float>, having both private val embedding: JsonField<List<Float>> and private val embeddingValue: JsonField<EmbeddingValue>? is unnecessary, no?

We can change the underlying data model while keeping backwards compat by implementing the existing methods in terms of the new data. For example, fun embeding() can be implemented as embeddingValue.getRequired("embedding").asFloatList()?

TomerAberbach · 2025-07-01T16:48:06Z

openai-java-core/src/main/kotlin/com/openai/models/embeddings/Embedding.kt

+            try {
+                this.embedding = JsonField.of(embedding.embedding().toMutableList())
+            } catch (e: Exception) {
+                // Fallback to field-level copying if embedding() method fails
+                this.embedding = embedding.embedding.map { it.toMutableList() }
+            }


I don't understand what this is for. Is this because we have the nullable and non-nullable fields? If so, then I guess this will go away once we implement my other suggestion?

TomerAberbach · 2025-07-01T16:51:25Z

openai-java-core/src/main/kotlin/com/openai/models/embeddings/EmbeddingCreateParams.kt

+            // Apply default encoding format if not explicitly set
+            if (body._encodingFormat().isMissing()) {
+                body.encodingFormat(EmbeddingDefaults.defaultEncodingFormat)
+            }


We shouldn't apply the default like this. We can just update the builder field to start out as:

private var encodingFormat: JsonField<EncodingFormat> = JsonField.of(EmbeddingDefaults.defaultEncodingFormat)

Then we also don't need that new internal method

TomerAberbach · 2025-07-01T16:52:07Z

openai-java-core/src/main/kotlin/com/openai/models/embeddings/EmbeddingValue.kt

+@JsonSerialize(using = EmbeddingValue.Serializer::class)
+class EmbeddingValue
+private constructor(
+    private val floatList: List<Float>? = null,


nit: here and throughout, we can call these floats instead of floatList

Suggested change

private val floatList: List<Float>? = null,

private val floats: List<Float>? = null,

TomerAberbach · 2025-07-01T16:52:41Z

openai-java-core/src/main/kotlin/com/openai/models/embeddings/EmbeddingValue.kt

+class EmbeddingValue
+private constructor(
+    private val floatList: List<Float>? = null,
+    private val base64String: String? = null,


nit: here and throughout, let's just call this base64. The type makes it clear it's a string

Suggested change

private val base64String: String? = null,

private val base64: String? = null,

TomerAberbach · 2025-07-01T16:55:03Z

openai-java-core/src/main/kotlin/com/openai/models/embeddings/EmbeddingDefaults.kt

This seems pretty overkill to me. I think we can just swap the default to base64 and people can set the encoding explicitly on the params object if they want floats over the wire?

In that case we can just delete this class and inline EncodingFormat.BASE64 in the params builder default

TomerAberbach · 2025-07-01T17:00:44Z

openai-java-core/src/main/kotlin/com/openai/models/embeddings/EmbeddingValue.kt

+            object : Visitor<Unit> {
+                override fun visitFloatList(floatList: List<Float>) {
+                    // Validate that float list is not empty and contains valid values
+                    if (floatList.isEmpty()) {
+                        throw OpenAIInvalidDataException("Float list cannot be empty")
+                    }
+                    floatList.forEach { value ->
+                        if (!value.isFinite()) {
+                            throw OpenAIInvalidDataException("Float values must be finite")
+                        }
+                    }
+                }
+
+                override fun visitBase64String(base64String: String) {
+                    // Validate base64 format
+                    try {
+                        Base64.getDecoder().decode(base64String)
+                    } catch (e: IllegalArgumentException) {
+                        throw OpenAIInvalidDataException("Invalid base64 string", e)
+                    }
+                }


In general the validate() methods in the SDK don't validate anything other than the "shape" of the data being correct (e.g. required fields are set)

In this case, I think we just want to check that the union is not a _json. So like:

accept( object : Visitor<Unit> { override fun visitFloatList(floatList: List<Float>) {} override fun visitBase64String(base64String: String) {} } )

Is sufficient. This will throw if it's a _json

TomerAberbach · 2025-07-01T17:06:03Z

openai-java-core/src/main/kotlin/com/openai/models/embeddings/EmbeddingValue.kt

+                }
+            }
+        )
+        return this // Return this instance if validation succeeds


We can just use apply like we do in other classes

TomerAberbach · 2025-07-01T17:06:43Z

openai-java-core/src/test/kotlin/com/openai/models/embeddings/EmbeddingDebugTest.kt

We should ultimately delete this since it's not really testing anything (there are no asserts)

Fine if you wanna use it to debug on your end

yoshioterada requested a review from a team as a code owner June 30, 2025 12:51

yoshioterada mentioned this pull request Jun 30, 2025

Fix Issue #211 : Improved Embedding Performance by Handling Base64 Encoding #303

Closed

Deleted invalid EmbeddingDefaults.kt

a7d881b

Deleted invalid EmbeddingDefaults.kt

TomerAberbach reviewed Jul 1, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Add EmbeddingValue union type and Base64 support for embeddings #519

Add EmbeddingValue union type and Base64 support for embeddings #519

Uh oh!

yoshioterada commented Jun 30, 2025

Uh oh!

yoshioterada commented Jun 30, 2025

Uh oh!

TomerAberbach Jul 1, 2025

Uh oh!

TomerAberbach Jul 1, 2025

Uh oh!

TomerAberbach Jul 1, 2025

Uh oh!

TomerAberbach Jul 1, 2025

Uh oh!

TomerAberbach Jul 1, 2025

Uh oh!

TomerAberbach Jul 1, 2025

Uh oh!

TomerAberbach Jul 1, 2025

Uh oh!

TomerAberbach Jul 1, 2025

Uh oh!

TomerAberbach Jul 1, 2025

Uh oh!

Uh oh!

	private val floatList: List<Float>? = null,
	private val floats: List<Float>? = null,

	private val base64String: String? = null,
	private val base64: String? = null,

Add EmbeddingValue union type and Base64 support for embeddings #519

Are you sure you want to change the base?

Add EmbeddingValue union type and Base64 support for embeddings #519

Uh oh!

Conversation

yoshioterada commented Jun 30, 2025

1. Default Behavior Change

Improvement from previous PR

Uh oh!

yoshioterada commented Jun 30, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!