Merge branch 'master' of github.com:Mintplex-Labs/anything-llm

server/utils/AiProviders/azureOpenAi/index.js

@@ -1,3 +1,5 @@
+const { toChunks } = require("../../helpers");
+
 class AzureOpenAi {
   constructor() {
     const { OpenAIClient, AzureKeyCredential } = require("@azure/openai");
@@ -6,6 +8,10 @@ class AzureOpenAi {
       new AzureKeyCredential(process.env.AZURE_OPENAI_KEY)
     );
     this.openai = openai;
+
+    // The maximum amount of "inputs" that OpenAI API can process in a single call.
+    // https://learn.microsoft.com/en-us/azure/ai-services/openai/faq#i-am-trying-to-use-embeddings-and-received-the-error--invalidrequesterror--too-many-inputs--the-max-number-of-inputs-is-1---how-do-i-fix-this-:~:text=consisting%20of%20up%20to%2016%20inputs%20per%20API%20request
+    this.embeddingChunkLimit = 16;
   }
 
   isValidChatModel(_modelName = "") {
@@ -83,10 +89,49 @@ class AzureOpenAi {
         "No EMBEDDING_MODEL_PREF ENV defined. This must the name of a deployment on your Azure account for an embedding model."
       );
 
-    const { data = [] } = await this.openai.getEmbeddings(
-      textEmbeddingModel,
-      textChunks
-    );
+    // Because there is a limit on how many chunks can be sent at once to Azure OpenAI
+    // we concurrently execute each max batch of text chunks possible.
+    // Refer to constructor embeddingChunkLimit for more info.
+    const embeddingRequests = [];
+    for (const chunk of toChunks(textChunks, this.embeddingChunkLimit)) {
+      embeddingRequests.push(
+        new Promise((resolve) => {
+          this.openai
+            .getEmbeddings(textEmbeddingModel, chunk)
+            .then((res) => {
+              resolve({ data: res.data, error: null });
+            })
+            .catch((e) => {
+              resolve({ data: [], error: e?.error });
+            });
+        })
+      );
+    }
+
+    const { data = [], error = null } = await Promise.all(
+      embeddingRequests
+    ).then((results) => {
+      // If any errors were returned from Azure abort the entire sequence because the embeddings
+      // will be incomplete.
+      const errors = results
+        .filter((res) => !!res.error)
+        .map((res) => res.error)
+        .flat();
+      if (errors.length > 0) {
+        return {
+          data: [],
+          error: `(${errors.length}) Embedding Errors! ${errors
+            .map((error) => `[${error.type}]: ${error.message}`)
+            .join(", ")}`,
+        };
+      }
+      return {
+        data: results.map((res) => res?.data || []).flat(),
+        error: null,
+      };
+    });
+
+    if (!!error) throw new Error(`Azure OpenAI Failed to embed: ${error}`);
     return data.length > 0 &&
       data.every((embd) => embd.hasOwnProperty("embedding"))
       ? data.map((embd) => embd.embedding)

server/utils/vectorDbProviders/pinecone/index.js

@@ -186,10 +186,12 @@ const Pinecone = {
     if (knownDocuments.length === 0) return;
 
     const vectorIds = knownDocuments.map((doc) => doc.vectorId);
-    await pineconeIndex.delete1({
-      ids: vectorIds,
-      namespace,
-    });
+    for (const batchOfVectorIds of toChunks(vectorIds, 1000)) {
+      await pineconeIndex.delete1({
+        ids: batchOfVectorIds,
+        namespace,
+      });
+    }
 
     const indexes = knownDocuments.map((doc) => doc.id);
     await DocumentVectors.deleteIds(indexes);