add support for new embeddings

Author: OkGoDoIt

OpenAI_API/Embedding/EmbeddingEndpoint.cs

@@ -9,9 +9,9 @@ namespace OpenAI_API.Embedding
 	public class EmbeddingEndpoint : EndpointBase, IEmbeddingEndpoint
 	{
 		/// <summary>
-		/// This allows you to send request to the recommended model without needing to specify. Every request uses the <see cref="Model.AdaTextEmbedding"/> model
+		/// This allows you to send request to a default model without needing to specify for each request.
 		/// </summary>
-		public EmbeddingRequest DefaultEmbeddingRequestArgs { get; set; } = new EmbeddingRequest() { Model = Model.AdaTextEmbedding };
+		public EmbeddingRequest DefaultEmbeddingRequestArgs { get; set; } = new EmbeddingRequest() { Model = Model.DefaultEmbeddingModel };
 
 		/// <summary>
 		/// The name of the endpoint, which is the final path segment in the API URL.  For example, "embeddings".
@@ -25,7 +25,7 @@ public class EmbeddingEndpoint : EndpointBase, IEmbeddingEndpoint
 		internal EmbeddingEndpoint(OpenAIAPI api) : base(api) { }
 
 		/// <summary>
-		/// Ask the API to embedd text using the default embedding model <see cref="Model.AdaTextEmbedding"/>
+		/// Ask the API to embed text using the default embedding model <see cref="Model.DefaultEmbeddingModel"/>
 		/// </summary>
 		/// <param name="input">Text to be embedded</param>
 		/// <returns>Asynchronously returns the embedding result. Look in its <see cref="Data.Embedding"/> property of <see cref="EmbeddingResult.Data"/> to find the vector of floating point numbers</returns>
@@ -36,7 +36,7 @@ public async Task<EmbeddingResult> CreateEmbeddingAsync(string input)
 		}
 
 		/// <summary>
-		/// Ask the API to embedd text using a custom request
+		/// Ask the API to embed text using a custom request
 		/// </summary>
 		/// <param name="request">Request to be send</param>
 		/// <returns>Asynchronously returns the embedding result. Look in its <see cref="Data.Embedding"/> property of <see cref="EmbeddingResult.Data"/> to find the vector of floating point numbers</returns>
@@ -45,16 +45,20 @@ public async Task<EmbeddingResult> CreateEmbeddingAsync(EmbeddingRequest request
 			return await HttpPost<EmbeddingResult>(postData: request);
 		}
 
-		/// <summary>
-		/// Ask the API to embedd text using the default embedding model <see cref="Model.AdaTextEmbedding"/>
-		/// </summary>
-		/// <param name="input">Text to be embedded</param>
-		/// <returns>Asynchronously returns the first embedding result as an array of floats.</returns>
+		/// <inheritdoc/>
 		public async Task<float[]> GetEmbeddingsAsync(string input)
 		{
 			EmbeddingRequest req = new EmbeddingRequest(DefaultEmbeddingRequestArgs.Model, input);
 			var embeddingResult = await CreateEmbeddingAsync(req);
 			return embeddingResult?.Data?[0]?.Embedding;
 		}
+
+		/// <inheritdoc/>
+		public async Task<float[]> GetEmbeddingsAsync(string input, Model model=null, int? dimensions = null)
+		{
+			EmbeddingRequest req = new EmbeddingRequest(model ?? Model.DefaultEmbeddingModel, input, dimensions);
+			var embeddingResult = await CreateEmbeddingAsync(req);
+			return embeddingResult?.Data?[0]?.Embedding;
+		}
 	}
 }

OpenAI_API/Embedding/EmbeddingRequest.cs

@@ -20,6 +20,12 @@ public class EmbeddingRequest
 		[JsonProperty("input")]
 		public string Input { get; set; }
 
+		/// <summary>
+		/// The number of dimensions the resulting output embeddings should have. Only supported in text-embedding-3 and later models.
+		/// </summary>
+		[JsonProperty("dimensions", NullValueHandling =NullValueHandling.Ignore)]
+		public int? Dimensions { get; set; }
+
 		/// <summary>
 		/// Cretes a new, empty <see cref="EmbeddingRequest"/>
 		/// </summary>
@@ -33,19 +39,21 @@ public EmbeddingRequest()
 		/// </summary>
 		/// <param name="model">The model to use. You can use <see cref="ModelsEndpoint.GetModelsAsync()"/> to see all of your available models, or use a standard model like <see cref="Model.AdaTextEmbedding"/>.</param>
 		/// <param name="input">The prompt to transform</param>
-		public EmbeddingRequest(Model model, string input)
+		/// <param name="dimensions">The number of dimensions the resulting output embeddings should have. Only supported in text-embedding-3 and later models.</param>
+		public EmbeddingRequest(Model model, string input, int? dimensions = null)
 		{
-			Model = model;
+			this.Model = model;
 			this.Input = input;
+			this.Dimensions = dimensions;
 		}
 
 		/// <summary>
-		/// Creates a new <see cref="EmbeddingRequest"/> with the specified input and the <see cref="Model.AdaTextEmbedding"/> model.
+		/// Creates a new <see cref="EmbeddingRequest"/> with the specified input and the <see cref="Model.DefaultEmbeddingModel"/> model.
 		/// </summary>
 		/// <param name="input">The prompt to transform</param>
 		public EmbeddingRequest(string input)
 		{
-			Model = OpenAI_API.Models.Model.AdaTextEmbedding;
+			this.Model = OpenAI_API.Models.Model.DefaultEmbeddingModel;
 			this.Input = input;
 		}
 	}

OpenAI_API/Embedding/IEmbeddingEndpoint.cs

@@ -8,30 +8,32 @@ namespace OpenAI_API.Embedding
 	/// </summary>
 	public interface IEmbeddingEndpoint
     {
-        /// <summary>
-        /// This allows you to send request to the recommended model without needing to specify. Every request uses the <see cref="Model.AdaTextEmbedding"/> model
-        /// </summary>
-        EmbeddingRequest DefaultEmbeddingRequestArgs { get; set; }
+		/// <summary>
+		/// This allows you to send request to a default model without needing to specify for each request
+		/// </summary>
+		EmbeddingRequest DefaultEmbeddingRequestArgs { get; set; }
 
-        /// <summary>
-        /// Ask the API to embedd text using the default embedding model <see cref="Model.AdaTextEmbedding"/>
-        /// </summary>
-        /// <param name="input">Text to be embedded</param>
-        /// <returns>Asynchronously returns the embedding result. Look in its <see cref="Data.Embedding"/> property of <see cref="EmbeddingResult.Data"/> to find the vector of floating point numbers</returns>
-        Task<EmbeddingResult> CreateEmbeddingAsync(string input);
+		/// <summary>
+		/// Ask the API to embed text using the default embedding model <see cref="Model.DefaultEmbeddingModel"/>
+		/// </summary>
+		/// <param name="input">Text to be embedded</param>
+		/// <returns>Asynchronously returns the embedding result. Look in its <see cref="Data.Embedding"/> property of <see cref="EmbeddingResult.Data"/> to find the vector of floating point numbers</returns>
+		Task<EmbeddingResult> CreateEmbeddingAsync(string input);
 
         /// <summary>
-        /// Ask the API to embedd text using a custom request
+        /// Ask the API to embed text using a custom request
         /// </summary>
         /// <param name="request">Request to be send</param>
         /// <returns>Asynchronously returns the embedding result. Look in its <see cref="Data.Embedding"/> property of <see cref="EmbeddingResult.Data"/> to find the vector of floating point numbers</returns>
         Task<EmbeddingResult> CreateEmbeddingAsync(EmbeddingRequest request);
 
-        /// <summary>
-        /// Ask the API to embedd text using the default embedding model <see cref="Model.AdaTextEmbedding"/>
-        /// </summary>
-        /// <param name="input">Text to be embedded</param>
-        /// <returns>Asynchronously returns the first embedding result as an array of floats.</returns>
-        Task<float[]> GetEmbeddingsAsync(string input);
+		/// <summary>
+		/// Ask the API to embed text <see cref="Model.DefaultEmbeddingModel"/>
+		/// </summary>
+		/// <param name="input">Text to be embedded</param>
+		/// <param name="model">The model to use. You can use <see cref="ModelsEndpoint.GetModelsAsync()"/> to see all of your available models, or use a standard model like <see cref="Model.AdaTextEmbedding"/>.</param>
+		/// <param name="dimensions">The number of dimensions the resulting output embeddings should have. Only supported in text-embedding-3 and later models.</param>
+		/// <returns>Asynchronously returns the first embedding result as an array of floats.</returns>
+		Task<float[]> GetEmbeddingsAsync(string input, Model model = null, int? dimensions = null);
     }
 }

OpenAI_API/Model/Model.cs

@@ -112,6 +112,11 @@ public Model()
 		/// </summary>
 		public static Model DefaultTranscriptionModel { get; set; } = Whisper1;
 
+		/// <summary>
+		/// The default model to use in embedding requests if no other model is specified.
+		/// </summary>
+		public static Model DefaultEmbeddingModel { get; set; } = AdaTextEmbedding;
+
 
 		/// <summary>
 		/// Gets more details about this Model from the API, specifically properties such as <see cref="OwnedBy"/> and permissions.
@@ -143,7 +148,7 @@ public async Task<Model> RetrieveModelDetailsAsync(OpenAI_API.OpenAIAPI api)
 		/// <summary>
 		///	The latest GPT-4 model with improved instruction following, JSON mode, reproducible outputs, parallel function calling, and more. Returns a maximum of 4,096 output tokens. This preview model is not yet suited for production traffic. 
 		/// </summary>
-		public static Model GPT4_Turbo => new Model("gpt-4-1106-preview") { OwnedBy = "openai" };
+		public static Model GPT4_Turbo => new Model("gpt-4-turbo-preview") { OwnedBy = "openai" };
 		#endregion
 
 		#region GPT-3.5
@@ -214,16 +219,27 @@ public async Task<Model> RetrieveModelDetailsAsync(OpenAI_API.OpenAIAPI api)
 		/// </summary>
 		[Obsolete("Will be deprecated by OpenAI on Jan 4th 2024.")]
 		public static Model DavinciCode => new Model("code-davinci-002") { OwnedBy = "openai" };
-#endregion
+		#endregion
 
-#region Embeddings
+		#region Embeddings
 		/// <summary>
-		/// OpenAI offers one second-generation embedding model for use with the embeddings API endpoint.
+		/// 2nd generation embedding model.
 		/// </summary>
 		public static Model AdaTextEmbedding => new Model("text-embedding-ada-002") { OwnedBy = "openai" };
-#endregion
 
-#region Moderation
+		/// <summary>
+		/// Most capable embedding model for both english and non-english tasks
+		/// </summary>
+		public static Model TextEmbedding3Large => new Model("text-embedding-3-large") { OwnedBy = "openai" };
+
+		/// <summary>
+		/// Increased performance over 2nd generation ada embedding model
+		/// </summary>
+		public static Model TextEmbedding3Small => new Model("text-embedding-3-small") { OwnedBy = "openai" };
+
+		#endregion
+
+		#region Moderation
 		/// <summary>
 		/// Stable text moderation model that may provide lower accuracy compared to TextModerationLatest.
 		/// OpenAI states they will provide advanced notice before updating this model.

OpenAI_Tests/ChatEndpointTests.cs

@@ -127,7 +127,7 @@ public void SimpleCompletion()
 
 		[TestCase("gpt-3.5-turbo")]
 		[TestCase("gpt-4")]
-		[TestCase("gpt-4-1106-preview")]
+		[TestCase("gpt-4-turbo-preview")]
 		public void ChatBackAndForth(string model)
 		{
 			var api = new OpenAI_API.OpenAIAPI();

OpenAI_Tests/EmbeddingEndpointTests.cs

@@ -6,63 +6,101 @@
 
 namespace OpenAI_Tests
 {
-    public class EmbeddingEndpointTests
-    {
-        [SetUp]
-        public void Setup()
-        {
-            OpenAI_API.APIAuthentication.Default = new OpenAI_API.APIAuthentication(Environment.GetEnvironmentVariable("TEST_OPENAI_SECRET_KEY"));
-        }
-
-        [Test]
-        public void GetBasicEmbedding()
-        {
-            var api = new OpenAI_API.OpenAIAPI();
-
-            Assert.IsNotNull(api.Embeddings);
-
-            var results = api.Embeddings.CreateEmbeddingAsync(new EmbeddingRequest(Model.AdaTextEmbedding, "A test text for embedding")).Result;
-            Assert.IsNotNull(results);
-            if (results.CreatedUnixTime.HasValue)
-            {
-                Assert.NotZero(results.CreatedUnixTime.Value);
-                Assert.NotNull(results.Created);
-                Assert.Greater(results.Created.Value, new DateTime(2018, 1, 1));
+	public class EmbeddingEndpointTests
+	{
+		[SetUp]
+		public void Setup()
+		{
+			OpenAI_API.APIAuthentication.Default = new OpenAI_API.APIAuthentication(Environment.GetEnvironmentVariable("TEST_OPENAI_SECRET_KEY"));
+		}
+
+		[Test]
+		public void GetBasicEmbedding()
+		{
+			var api = new OpenAI_API.OpenAIAPI();
+
+			Assert.IsNotNull(api.Embeddings);
+
+			var results = api.Embeddings.CreateEmbeddingAsync(new EmbeddingRequest(Model.AdaTextEmbedding, "A test text for embedding")).Result;
+			Assert.IsNotNull(results);
+			if (results.CreatedUnixTime.HasValue)
+			{
+				Assert.NotZero(results.CreatedUnixTime.Value);
+				Assert.NotNull(results.Created);
+				Assert.Greater(results.Created.Value, new DateTime(2018, 1, 1));
 				Assert.Less(results.Created.Value, DateTime.Now.AddDays(1));
-			} else
-            {
-                Assert.Null(results.Created);
 			}
-            Assert.NotNull(results.Object);
-            Assert.NotZero(results.Data.Count);
-            Assert.That(results.Data.First().Embedding.Length == 1536);
-        }
+			else
+			{
+				Assert.Null(results.Created);
+			}
+			Assert.NotNull(results.Object);
+			Assert.NotZero(results.Data.Count);
+			Assert.That(results.Data.First().Embedding.Length == 1536);
+		}
 
-        [Test]
-        public void ReturnedUsage()
-        {
-            var api = new OpenAI_API.OpenAIAPI();
+		[Test]
+		public void ReturnedUsage()
+		{
+			var api = new OpenAI_API.OpenAIAPI();
 
-            Assert.IsNotNull(api.Embeddings);
+			Assert.IsNotNull(api.Embeddings);
 
-            var results = api.Embeddings.CreateEmbeddingAsync(new EmbeddingRequest(Model.AdaTextEmbedding, "A test text for embedding")).Result;
-            Assert.IsNotNull(results);
+			var results = api.Embeddings.CreateEmbeddingAsync(new EmbeddingRequest(Model.AdaTextEmbedding, "A test text for embedding")).Result;
+			Assert.IsNotNull(results);
 
 			Assert.IsNotNull(results.Usage);
 			Assert.GreaterOrEqual(results.Usage.PromptTokens, 5);
 			Assert.GreaterOrEqual(results.Usage.TotalTokens, results.Usage.PromptTokens);
 		}
 
-        [Test]
-        public void GetSimpleEmbedding()
-        {
-            var api = new OpenAI_API.OpenAIAPI();
+		[Test]
+		public void GetSimpleEmbedding()
+		{
+			var api = new OpenAI_API.OpenAIAPI();
+
+			Assert.IsNotNull(api.Embeddings);
+
+			var results = api.Embeddings.GetEmbeddingsAsync("A test text for embedding").Result;
+			Assert.IsNotNull(results);
+			Assert.That(results.Length == 1536);
+		}
+
+		[Test]
+		public void GetEmbeddingWithLargeModel()
+		{
+			var api = new OpenAI_API.OpenAIAPI();
 
-            Assert.IsNotNull(api.Embeddings);
+			Assert.IsNotNull(api.Embeddings);
 
-            var results = api.Embeddings.GetEmbeddingsAsync("A test text for embedding").Result;
-            Assert.IsNotNull(results);
-            Assert.That(results.Length == 1536);
-        }
-    }
+			var results = api.Embeddings.GetEmbeddingsAsync("A test text for embedding", Model.TextEmbedding3Large).Result;
+			Assert.IsNotNull(results);
+			Assert.That(results.Length == 3072);
+		}
+
+		[Test]
+		public void GetEmbeddingWithSmallModel()
+		{
+			var api = new OpenAI_API.OpenAIAPI();
+
+			Assert.IsNotNull(api.Embeddings);
+
+			var results = api.Embeddings.GetEmbeddingsAsync("A test text for embedding", Model.TextEmbedding3Small).Result;
+			Assert.IsNotNull(results);
+			Assert.That(results.Length == 1536);
+		}
+
+
+		[Test]
+		public void GetEmbeddingWithDimensions()
+		{
+			var api = new OpenAI_API.OpenAIAPI();
+
+			Assert.IsNotNull(api.Embeddings);
+
+			var results = api.Embeddings.GetEmbeddingsAsync("A test text for embedding", Model.TextEmbedding3Small, 350).Result;
+			Assert.IsNotNull(results);
+			Assert.That(results.Length == 350);
+		}
+	}
 }