Is it possible to allow changes to the `encoding_format` of the embedded options?

[JadynWong]

Currently EmbeddingClient fixes the encoding_format value to base64 for better performance.

openai-dotnet/src/Custom/Embeddings/EmbeddingGenerationOptions.cs

Lines 77 to 82 in 45fc4d7

	// CUSTOM: Made internal. We always request the embedding as a base64-encoded string for better performance.
	/// <summary>
	/// The format to return the embeddings in. Can be either `float` or
	/// [`base64`](https://pypi.org/project/pybase64/).
	/// </summary>
	internal InternalEmbeddingGenerationOptionsEncodingFormat? EncodingFormat { get; set; }

openai-dotnet/src/Custom/Embeddings/Embedding.cs

Lines 75 to 84 in 45fc4d7

	internal Embedding(int index, BinaryData embeddingProperty, InternalEmbeddingObject @object, IDictionary<string, BinaryData> serializedAdditionalRawData)
	{
	Index = (int)index;
	EmbeddingProperty = embeddingProperty;
	Object = @object;
	_serializedAdditionalRawData = serializedAdditionalRawData;
	// Handle additional custom properties.
	Vector = ConvertToVectorOfFloats(embeddingProperty);
	}

It can't be changed, even if I want to use float format. I want to use this client for text-embeddings-inference, which currently does not support the encoding_format parameter.
This results in the following error

The input is not a valid Base64 string of encoded floats.

I know that encoding_format compatibility would be a better approach in other projects, but a lot of compatible openai api's don't update as fast as they should.

Is it possible to allow users to change the encoding_format value?
Of course, as the official SDK of OpenAI, I would respect it if it was only compatible with OpenAI.

For now I can serialize it myself using protocol methods.

[joseharriaga]

Thank you for reaching out, @JadynWong ! Presently, it is not in our plans to expose the encoding_format property publicly, but we will definitely take your feedback into consideration as we continue to evolve this API.

As you have correctly pointed out, you can use the EmbeddingClient's GenerateEmbeddings protocol method to have full control of the request that is sent to the OpenAI service. By the sound of it, you have it all figured out already, but I am including a small example below in case others might find it useful:

EmbeddingClient client = new("text-embedding-3-small", Environment.GetEnvironmentVariable("OPENAI_API_KEY"));

string description = "Best hotel in town if you like luxury hotels. They have an amazing infinity pool, a spa,"
    + " and a really helpful concierge. The location is perfect -- right downtown, close to all the tourist"
    + " attractions. We highly recommend this hotel.";

BinaryData input = BinaryData.FromObjectAsJson(new
{
    model = "text-embedding-3-small",
    input = description,
    encoding_format = "float"
});

using BinaryContent content = BinaryContent.Create(input);
ClientResult result = await client.GenerateEmbeddingsAsync(content);
BinaryData output = result.GetRawResponse().Content;

using JsonDocument outputAsJson = JsonDocument.Parse(output.ToString());
JsonElement vector = outputAsJson.RootElement
    .GetProperty("data"u8)[0]
    .GetProperty("embedding"u8);

Console.WriteLine($"Dimension: {vector.GetArrayLength()}");
Console.WriteLine($"Floats: ");
int i = 0;
foreach (JsonElement element in vector.EnumerateArray())
{
    Console.WriteLine($"  [{i++,4}] = {element.GetDouble()}");
}

Note that, when using protocol methods, the input and output (represented as BinarayData) map directly to the REST API as described here: 🔗 https://platform.openai.com/docs/api-reference/embeddings/create

I have also created the following PR to add this example to our suite: 🔗 #122