fix for OpenAI to do embeddings in batches

src/libs/Providers/LangChain.Providers.OpenAI/OpenAiModel.Embeddings.cs

@@ -1,6 +1,7 @@
 using System.Diagnostics;
 using OpenAI.Constants;
 using OpenAI.Embeddings;
+using static System.Net.Mime.MediaTypeNames;
 
 namespace LangChain.Providers.OpenAI;
 
@@ -57,14 +58,46 @@ public async Task<float[]> EmbedQueryAsync(
         return response.Data[0].Embedding.Select(static x => (float)x).ToArray();
     }
 
+    const int MaxElementsPerRequest = 2048;
     /// <inheritdoc/>
     public async Task<float[][]> EmbedDocumentsAsync(
         string[] texts,
         CancellationToken cancellationToken = default)
     {
-        return await Task.WhenAll(
-            texts
-                .Select(text => EmbedQueryAsync(text, cancellationToken))).ConfigureAwait(false);
+
+        // API has limit of 2048 elements in array per request
+        // so we need to split texts into batches
+        // https://platform.openai.com/docs/api-reference/embeddings
+
+
+        var watch = Stopwatch.StartNew();
+        List<float[]> result = new();
+        List<string> allTexts = new(texts);
+
+        while (allTexts.Count > 0)
+        {
+            var currentBatch = allTexts.Take(MaxElementsPerRequest).ToArray();
+            allTexts.RemoveRange(0, currentBatch.Length);
+
+            var response = await Api.EmbeddingsEndpoint.CreateEmbeddingAsync(
+                request: new EmbeddingsRequest(
+                    input: currentBatch,
+                    model: EmbeddingModelId,
+                    user: User),
+                cancellationToken).ConfigureAwait(false);
+
+            var usage = GetUsage(response) with
+            {
+                Time = watch.Elapsed,
+            };
+            lock (_usageLock)
+            {
+                TotalUsage += usage;
+            }
+            result.AddRange(response.Data.Select(static x => x.Embedding.Select(static x => (float)x).ToArray()));
+        }
+
+        return result.ToArray();
     }
 
     #endregion