groq[minor]: Fix streaming metadata back to client (langchain-ai#6573)

* groq[minor]: Fix streaming metadata back to client * chore: lint files * chore: lint files * implemented usage metadata for invoke too
CarterMorris · Nov 10, 2024 · 04a9c7d · 04a9c7d
1 parent a6c7b5a
commit 04a9c7d
Show file tree

Hide file tree

Showing 2 changed files with 48 additions and 20 deletions.
diff --git a/libs/langchain-groq/src/chat_models.ts b/libs/langchain-groq/src/chat_models.ts
@@ -23,6 +23,7 @@ import {
   OpenAIToolCall,
   isAIMessage,
   BaseMessageChunk,
+  UsageMetadata,
 } from "@langchain/core/messages";
 import {
   ChatGeneration,
@@ -179,7 +180,8 @@ function convertMessagesToGroqParams(
 }
 
 function groqResponseToChatMessage(
-  message: ChatCompletionsAPI.ChatCompletionMessage
+  message: ChatCompletionsAPI.ChatCompletionMessage,
+  usageMetadata?: UsageMetadata
 ): BaseMessage {
   const rawToolCalls: OpenAIToolCall[] | undefined = message.tool_calls as
     | OpenAIToolCall[]
@@ -201,6 +203,7 @@ function groqResponseToChatMessage(
         additional_kwargs: { tool_calls: rawToolCalls },
         tool_calls: toolCalls,
         invalid_tool_calls: invalidToolCalls,
+        usage_metadata: usageMetadata,
       });
     }
     default:
@@ -226,7 +229,8 @@ function _convertDeltaToolCallToToolCallChunk(
 function _convertDeltaToMessageChunk(
   // eslint-disable-next-line @typescript-eslint/no-explicit-any
   delta: Record<string, any>,
-  index: number
+  index: number,
+  xGroq?: ChatCompletionsAPI.ChatCompletionChunk.XGroq
 ): {
   message: BaseMessageChunk;
   toolCallData?: {
@@ -250,6 +254,18 @@ function _convertDeltaToMessageChunk(
   } else {
     additional_kwargs = {};
   }
+
+  let usageMetadata: UsageMetadata | undefined;
+  let groqMessageId: string | undefined;
+  if (xGroq?.usage) {
+    usageMetadata = {
+      input_tokens: xGroq.usage.prompt_tokens,
+      output_tokens: xGroq.usage.completion_tokens,
+      total_tokens: xGroq.usage.total_tokens,
+    };
+    groqMessageId = xGroq.id;
+  }
+
   if (role === "user") {
     return {
       message: new HumanMessageChunk({ content }),
@@ -270,6 +286,8 @@ function _convertDeltaToMessageChunk(
               index: tc.index,
             }))
           : undefined,
+        usage_metadata: usageMetadata,
+        id: groqMessageId,
       }),
       toolCallData: toolCallChunks
         ? toolCallChunks.map((tc) => ({
@@ -771,7 +789,10 @@ export class ChatGroq extends BaseChatModel<
       index: number;
       type: "tool_call_chunk";
     }[] = [];
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    let responseMetadata: Record<string, any> | undefined;
     for await (const data of response) {
+      responseMetadata = data;
       const choice = data?.choices[0];
       if (!choice) {
         continue;
@@ -787,7 +808,8 @@ export class ChatGroq extends BaseChatModel<
           ...choice.delta,
           role,
         } ?? {},
-        choice.index
+        choice.index,
+        data.x_groq
       );
 
       if (toolCallData) {
@@ -818,6 +840,19 @@ export class ChatGroq extends BaseChatModel<
       void runManager?.handleLLMNewToken(chunk.text ?? "");
     }
 
+    if (responseMetadata) {
+      if ("choices" in responseMetadata) {
+        delete responseMetadata.choices;
+      }
+      yield new ChatGenerationChunk({
+        message: new AIMessageChunk({
+          content: "",
+          response_metadata: responseMetadata,
+        }),
+        text: "",
+      });
+    }
+
     if (options.signal?.aborted) {
       throw new Error("AbortError");
     }
@@ -898,10 +933,19 @@ export class ChatGroq extends BaseChatModel<
     if ("choices" in data && data.choices) {
       for (const part of (data as ChatCompletion).choices) {
         const text = part.message?.content ?? "";
+        let usageMetadata: UsageMetadata | undefined;
+        if (tokenUsage.totalTokens !== undefined) {
+          usageMetadata = {
+            input_tokens: tokenUsage.promptTokens ?? 0,
+            output_tokens: tokenUsage.completionTokens ?? 0,
+            total_tokens: tokenUsage.totalTokens,
+          };
+        }
         const generation: ChatGeneration = {
           text,
           message: groqResponseToChatMessage(
-            part.message ?? { role: "assistant" }
+            part.message ?? { role: "assistant" },
+            usageMetadata
           ),
         };
         generation.generationInfo = {

diff --git a/libs/langchain-groq/src/tests/chat_models.standard.int.test.ts b/libs/langchain-groq/src/tests/chat_models.standard.int.test.ts
@@ -25,22 +25,6 @@ class ChatGroqStandardIntegrationTests extends ChatModelIntegrationTests<
     });
   }
 
-  async testUsageMetadataStreaming() {
-    this.skipTestMessage(
-      "testUsageMetadataStreaming",
-      "ChatGroq",
-      "Streaming tokens is not currently supported."
-    );
-  }
-
-  async testUsageMetadata() {
-    this.skipTestMessage(
-      "testUsageMetadata",
-      "ChatGroq",
-      "Usage metadata tokens is not currently supported."
-    );
-  }
-
   async testToolMessageHistoriesListContent() {
     this.skipTestMessage(
       "testToolMessageHistoriesListContent",