@langchain/google-genai 0.2.17 → 0.2.18

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (3) hide show
  1. package/dist/chat_models.cjs +21 -19
  2. package/dist/chat_models.js +21 -19
  3. package/package.json +1 -1
package/dist/chat_models.cjs CHANGED
@@ -697,30 +697,32 @@ class ChatGoogleGenerativeAI extends chat_models_1.BaseChatModel {
697
697
  return stream;
698
698
  });
699
699
  let usageMetadata;
700
+ // Keep prior cumulative counts for calculating token deltas while streaming
701
+ let prevPromptTokenCount = 0;
702
+ let prevCandidatesTokenCount = 0;
703
+ let prevTotalTokenCount = 0;
700
704
  let index = 0;
701
705
  for await (const response of stream) {
702
706
  if ("usageMetadata" in response &&
707
+ response.usageMetadata !== undefined &&
703
708
  this.streamUsage !== false &&
704
709
  options.streamUsage !== false) {
705
- const genAIUsageMetadata = response.usageMetadata;
706
- if (!usageMetadata) {
707
- usageMetadata = {
708
- input_tokens: genAIUsageMetadata.promptTokenCount ?? 0,
709
- output_tokens: genAIUsageMetadata.candidatesTokenCount ?? 0,
710
- total_tokens: genAIUsageMetadata.totalTokenCount ?? 0,
711
- };
712
- }
713
- else {
714
- // Under the hood, LangChain combines the prompt tokens. Google returns the updated
715
- // total each time, so we need to find the difference between the tokens.
716
- const outputTokenDiff = (genAIUsageMetadata.candidatesTokenCount ?? 0) -
717
- usageMetadata.output_tokens;
718
- usageMetadata = {
719
- input_tokens: 0,
720
- output_tokens: outputTokenDiff,
721
- total_tokens: outputTokenDiff,
722
- };
723
- }
710
+ usageMetadata = {
711
+ input_tokens: response.usageMetadata.promptTokenCount ?? 0,
712
+ output_tokens: response.usageMetadata.candidatesTokenCount ?? 0,
713
+ total_tokens: response.usageMetadata.totalTokenCount ?? 0,
714
+ };
715
+ // Under the hood, LangChain combines the prompt tokens. Google returns the updated
716
+ // total each time, so we need to find the difference between the tokens.
717
+ const newPromptTokenCount = response.usageMetadata.promptTokenCount ?? 0;
718
+ usageMetadata.input_tokens = Math.max(0, newPromptTokenCount - prevPromptTokenCount);
719
+ prevPromptTokenCount = newPromptTokenCount;
720
+ const newCandidatesTokenCount = response.usageMetadata.candidatesTokenCount ?? 0;
721
+ usageMetadata.output_tokens = Math.max(0, newCandidatesTokenCount - prevCandidatesTokenCount);
722
+ prevCandidatesTokenCount = newCandidatesTokenCount;
723
+ const newTotalTokenCount = response.usageMetadata.totalTokenCount ?? 0;
724
+ usageMetadata.total_tokens = Math.max(0, newTotalTokenCount - prevTotalTokenCount);
725
+ prevTotalTokenCount = newTotalTokenCount;
724
726
  }
725
727
  const chunk = (0, common_js_1.convertResponseContentToChatGenerationChunk)(response, {
726
728
  usageMetadata,
package/dist/chat_models.js CHANGED
@@ -694,30 +694,32 @@ export class ChatGoogleGenerativeAI extends BaseChatModel {
694
694
  return stream;
695
695
  });
696
696
  let usageMetadata;
697
+ // Keep prior cumulative counts for calculating token deltas while streaming
698
+ let prevPromptTokenCount = 0;
699
+ let prevCandidatesTokenCount = 0;
700
+ let prevTotalTokenCount = 0;
697
701
  let index = 0;
698
702
  for await (const response of stream) {
699
703
  if ("usageMetadata" in response &&
704
+ response.usageMetadata !== undefined &&
700
705
  this.streamUsage !== false &&
701
706
  options.streamUsage !== false) {
702
- const genAIUsageMetadata = response.usageMetadata;
703
- if (!usageMetadata) {
704
- usageMetadata = {
705
- input_tokens: genAIUsageMetadata.promptTokenCount ?? 0,
706
- output_tokens: genAIUsageMetadata.candidatesTokenCount ?? 0,
707
- total_tokens: genAIUsageMetadata.totalTokenCount ?? 0,
708
- };
709
- }
710
- else {
711
- // Under the hood, LangChain combines the prompt tokens. Google returns the updated
712
- // total each time, so we need to find the difference between the tokens.
713
- const outputTokenDiff = (genAIUsageMetadata.candidatesTokenCount ?? 0) -
714
- usageMetadata.output_tokens;
715
- usageMetadata = {
716
- input_tokens: 0,
717
- output_tokens: outputTokenDiff,
718
- total_tokens: outputTokenDiff,
719
- };
720
- }
707
+ usageMetadata = {
708
+ input_tokens: response.usageMetadata.promptTokenCount ?? 0,
709
+ output_tokens: response.usageMetadata.candidatesTokenCount ?? 0,
710
+ total_tokens: response.usageMetadata.totalTokenCount ?? 0,
711
+ };
712
+ // Under the hood, LangChain combines the prompt tokens. Google returns the updated
713
+ // total each time, so we need to find the difference between the tokens.
714
+ const newPromptTokenCount = response.usageMetadata.promptTokenCount ?? 0;
715
+ usageMetadata.input_tokens = Math.max(0, newPromptTokenCount - prevPromptTokenCount);
716
+ prevPromptTokenCount = newPromptTokenCount;
717
+ const newCandidatesTokenCount = response.usageMetadata.candidatesTokenCount ?? 0;
718
+ usageMetadata.output_tokens = Math.max(0, newCandidatesTokenCount - prevCandidatesTokenCount);
719
+ prevCandidatesTokenCount = newCandidatesTokenCount;
720
+ const newTotalTokenCount = response.usageMetadata.totalTokenCount ?? 0;
721
+ usageMetadata.total_tokens = Math.max(0, newTotalTokenCount - prevTotalTokenCount);
722
+ prevTotalTokenCount = newTotalTokenCount;
721
723
  }
722
724
  const chunk = convertResponseContentToChatGenerationChunk(response, {
723
725
  usageMetadata,
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@langchain/google-genai",
3
- "version": "0.2.17",
3
+ "version": "0.2.18",
4
4
  "description": "Google Generative AI integration for LangChain.js",
5
5
  "type": "module",
6
6
  "engines": {