V4.8.17 feature (#3493)

* split tokens into input and output (#3477) * split tokens into input and output * query extension & tool call & question guide * fix * perf: input and output tokens * perf: tool call if else * perf: remove code * fix: extract usage count * fix: qa usage count --------- Co-authored-by: heheer <heheer@sealos.io>
2025-07-23 13:03:50 +00:00 · 2024-12-30 10:13:25 +08:00
parent da2831b948
commit 50bf7f9a3b
46 changed files with 467 additions and 230 deletions
--- a/projects/app/src/components/core/ai/AISettingModal/index.tsx
+++ b/projects/app/src/components/core/ai/AISettingModal/index.tsx
@@ -18,7 +18,6 @@ import {
  Thead,
  Tr,
  Table,
-  useDisclosure,
  FlexProps
 } from '@chakra-ui/react';
 import { useSystemStore } from '@/web/common/system/useSystemStore';
@@ -175,10 +174,28 @@ const AIChatSettingsModal = ({
            <Tbody>
              <Tr color={'myGray.900'}>
                <Td pt={0} pb={2}>
-                  {t('common:support.wallet.Ai point every thousand tokens', {
-                    points: selectedModel?.charsPointsPrice || 0
-                  })}
+                  {typeof selectedModel?.inputPrice === 'number' ? (
+                    <>
+                      <Box>
+                        {t('common:support.wallet.Ai point every thousand tokens_input', {
+                          points: selectedModel?.inputPrice || 0
+                        })}
+                      </Box>
+                      <Box>
+                        {t('common:support.wallet.Ai point every thousand tokens_output', {
+                          points: selectedModel?.outputPrice || 0
+                        })}
+                      </Box>
+                    </>
+                  ) : (
+                    <>
+                      {t('common:support.wallet.Ai point every thousand tokens', {
+                        points: selectedModel?.charsPointsPrice || 0
+                      })}
+                    </>
+                  )}
                </Td>
+
                <Td pt={0} pb={2}>
                  {Math.round((selectedModel?.maxContext || 4096) / 1000)}K
                </Td>
--- a/projects/app/src/components/core/ai/ModelTable/index.tsx
+++ b/projects/app/src/components/core/ai/ModelTable/index.tsx
@@ -60,14 +60,32 @@ const ModelTable = () => {
    const formatLLMModelList = llmModelList.map((item) => ({
      ...item,
      typeLabel: t('common:model.type.chat'),
-      priceLabel: (
-        <Flex color={'myGray.700'}>
-          <Box fontWeight={'bold'} color={'myGray.900'} mr={0.5}>
-            {item.charsPointsPrice}
+      priceLabel:
+        typeof item.inputPrice === 'number' ? (
+          <Box>
+            <Flex>
+              {`${t('common:common.Input')}:`}
+              <Box fontWeight={'bold'} color={'myGray.900'} mr={0.5} ml={2}>
+                {item.inputPrice || 0}
+              </Box>
+              {`${t('common:support.wallet.subscription.point')} / 1K Tokens`}
+            </Flex>
+            <Flex>
+              {`${t('common:common.Output')}:`}
+              <Box fontWeight={'bold'} color={'myGray.900'} mr={0.5} ml={2}>
+                {item.outputPrice || 0}
+              </Box>
+              {`${t('common:support.wallet.subscription.point')} / 1K Tokens`}
+            </Flex>
          </Box>
-          {`${t('common:support.wallet.subscription.point')} / 1K Tokens`}
-        </Flex>
-      ),
+        ) : (
+          <Flex color={'myGray.700'}>
+            <Box fontWeight={'bold'} color={'myGray.900'} mr={0.5}>
+              {item.charsPointsPrice}
+            </Box>
+            {`${t('common:support.wallet.subscription.point')} / 1K Tokens`}
+          </Flex>
+        ),
      tagColor: 'blue'
    }));
    const formatVectorModelList = vectorModelList.map((item) => ({
@@ -149,13 +167,13 @@ const ModelTable = () => {

    return filterList;
  }, [
-    provider,
-    modelType,
    llmModelList,
    vectorModelList,
    audioSpeechModelList,
    whisperModel,
    t,
+    modelType,
+    provider,
    search
  ]);

--- a/projects/app/src/components/core/chat/components/WholeResponseModal.tsx
+++ b/projects/app/src/components/core/chat/components/WholeResponseModal.tsx
@@ -155,10 +155,26 @@ export const WholeResponseContent = ({
          label={t('common:core.chat.response.module tokens')}
          value={`${activeModule?.tokens}`}
        />
+        <Row
+          label={t('common:core.chat.response.module input tokens')}
+          value={`${activeModule?.inputTokens}`}
+        />
+        <Row
+          label={t('common:core.chat.response.module output tokens')}
+          value={`${activeModule?.outputTokens}`}
+        />
        <Row
          label={t('common:core.chat.response.Tool call tokens')}
          value={`${activeModule?.toolCallTokens}`}
        />
+        <Row
+          label={t('common:core.chat.response.Tool call input tokens')}
+          value={`${activeModule?.toolCallInputTokens}`}
+        />
+        <Row
+          label={t('common:core.chat.response.Tool call output tokens')}
+          value={`${activeModule?.toolCallOutputTokens}`}
+        />

        <Row label={t('common:core.chat.response.module query')} value={activeModule?.query} />
        <Row
--- a/projects/app/src/pages/account/usage/UsageDetail.tsx
+++ b/projects/app/src/pages/account/usage/UsageDetail.tsx
@@ -26,37 +26,48 @@ const UsageDetail = ({ usage, onClose }: { usage: UsageItemType; onClose: () =>
    [usage.list]
  );

-  const { hasModel, hasToken, hasCharsLen, hasDuration } = useMemo(() => {
-    let hasModel = false;
-    let hasToken = false;
-    let hasCharsLen = false;
-    let hasDuration = false;
-    let hasDataLen = false;
+  const { hasModel, hasToken, hasInputToken, hasOutputToken, hasCharsLen, hasDuration } =
+    useMemo(() => {
+      let hasModel = false;
+      let hasToken = false;
+      let hasInputToken = false;
+      let hasOutputToken = false;
+      let hasCharsLen = false;
+      let hasDuration = false;
+      let hasDataLen = false;

-    usage.list.forEach((item) => {
-      if (item.model !== undefined) {
-        hasModel = true;
-      }
+      usage.list.forEach((item) => {
+        if (item.model !== undefined) {
+          hasModel = true;
+        }

-      if (typeof item.tokens === 'number') {
-        hasToken = true;
-      }
-      if (typeof item.charsLength === 'number') {
-        hasCharsLen = true;
-      }
-      if (typeof item.duration === 'number') {
-        hasDuration = true;
-      }
-    });
+        if (typeof item.tokens === 'number') {
+          hasToken = true;
+        }
+        if (typeof item.inputTokens === 'number') {
+          hasInputToken = true;
+        }
+        if (typeof item.outputTokens === 'number') {
+          hasOutputToken = true;
+        }
+        if (typeof item.charsLength === 'number') {
+          hasCharsLen = true;
+        }
+        if (typeof item.duration === 'number') {
+          hasDuration = true;
+        }
+      });

-    return {
-      hasModel,
-      hasToken,
-      hasCharsLen,
-      hasDuration,
-      hasDataLen
-    };
-  }, [usage.list]);
+      return {
+        hasModel,
+        hasToken,
+        hasInputToken,
+        hasOutputToken,
+        hasCharsLen,
+        hasDuration,
+        hasDataLen
+      };
+    }, [usage.list]);

  return (
    <MyModal
@@ -98,6 +109,8 @@ const UsageDetail = ({ usage, onClose }: { usage: UsageItemType; onClose: () =>
                  <Th>{t('account_usage:module_name')}</Th>
                  {hasModel && <Th>{t('account_usage:ai_model')}</Th>}
                  {hasToken && <Th>{t('account_usage:token_length')}</Th>}
+                  {hasInputToken && <Th>{t('account_usage:input_token_length')}</Th>}
+                  {hasOutputToken && <Th>{t('account_usage:output_token_length')}</Th>}
                  {hasCharsLen && <Th>{t('account_usage:text_length')}</Th>}
                  {hasDuration && <Th>{t('account_usage:duration_seconds')}</Th>}
                  <Th>{t('account_usage:total_points_consumed')}</Th>
@@ -109,6 +122,8 @@ const UsageDetail = ({ usage, onClose }: { usage: UsageItemType; onClose: () =>
                    <Td>{t(item.moduleName as any)}</Td>
                    {hasModel && <Td>{item.model ?? '-'}</Td>}
                    {hasToken && <Td>{item.tokens ?? '-'}</Td>}
+                    {hasInputToken && <Td>{item.inputTokens ?? '-'}</Td>}
+                    {hasOutputToken && <Td>{item.outputTokens ?? '-'}</Td>}
                    {hasCharsLen && <Td>{item.charsLength ?? '-'}</Td>}
                    {hasDuration && <Td>{item.duration ?? '-'}</Td>}
                    <Td>{formatNumber(item.amount)}</Td>
--- a/projects/app/src/pages/api/core/ai/agent/createQuestionGuide.ts
+++ b/projects/app/src/pages/api/core/ai/agent/createQuestionGuide.ts
@@ -37,7 +37,7 @@ async function handler(

    const qgModel = global.llmModels[0];

-    const { result, tokens } = await createQuestionGuide({
+    const { result, inputTokens, outputTokens } = await createQuestionGuide({
      messages,
      model: qgModel.model
    });
@@ -47,7 +47,8 @@ async function handler(
    });

    pushQuestionGuideUsage({
-      tokens,
+      inputTokens,
+      outputTokens,
      teamId,
      tmbId
    });
--- a/projects/app/src/pages/api/core/ai/agent/v2/createQuestionGuide.ts
+++ b/projects/app/src/pages/api/core/ai/agent/v2/createQuestionGuide.ts
@@ -52,14 +52,15 @@ async function handler(req: ApiRequestProps<CreateQuestionGuideParams>, res: Nex

  const qgModel = questionGuide?.model || global.llmModels[0].model;

-  const { result, tokens } = await createQuestionGuide({
+  const { result, inputTokens, outputTokens } = await createQuestionGuide({
    messages,
    model: qgModel,
    customPrompt: questionGuide?.customPrompt
  });

  pushQuestionGuideUsage({
-    tokens,
+    inputTokens,
+    outputTokens,
    teamId,
    tmbId
  });
--- a/projects/app/src/pages/api/core/dataset/data/insertData.ts
+++ b/projects/app/src/pages/api/core/dataset/data/insertData.ts
@@ -89,7 +89,7 @@ async function handler(req: NextApiRequest) {
  pushGenerateVectorUsage({
    teamId,
    tmbId,
-    tokens,
+    inputTokens: tokens,
    model: vectorModelData.model
  });

--- a/projects/app/src/pages/api/core/dataset/data/update.ts
+++ b/projects/app/src/pages/api/core/dataset/data/update.ts
@@ -36,7 +36,7 @@ async function handler(req: ApiRequestProps<UpdateDatasetDataProps>) {
    pushGenerateVectorUsage({
      teamId,
      tmbId,
-      tokens,
+      inputTokens: tokens,
      model: vectorModel
    });
  } else {
--- a/projects/app/src/pages/api/core/dataset/searchTest.ts
+++ b/projects/app/src/pages/api/core/dataset/searchTest.ts
@@ -74,14 +74,15 @@ async function handler(req: NextApiRequest) {
  const { totalPoints } = pushGenerateVectorUsage({
    teamId,
    tmbId,
-    tokens,
+    inputTokens: tokens,
    model: dataset.vectorModel,
    source: apikey ? UsageSourceEnum.api : UsageSourceEnum.fastgpt,

    ...(aiExtensionResult &&
      extensionModel && {
        extensionModel: extensionModel.name,
-        extensionTokens: aiExtensionResult.tokens
+        extensionInputTokens: aiExtensionResult.inputTokens,
+        extensionOutputTokens: aiExtensionResult.outputTokens
      })
  });
  if (apikey) {
--- a/projects/app/src/pages/api/v1/embeddings.ts
+++ b/projects/app/src/pages/api/v1/embeddings.ts
@@ -57,7 +57,7 @@ async function handler(req: NextApiRequest, res: NextApiResponse<any>) {
  const { totalPoints } = pushGenerateVectorUsage({
    teamId,
    tmbId,
-    tokens,
+    inputTokens: tokens,
    model,
    billId,
    source: getUsageSourceByAuthType({ authType })
--- a/projects/app/src/pages/dataset/detail/components/Import/Context.tsx
+++ b/projects/app/src/pages/dataset/detail/components/Import/Context.tsx
@@ -196,7 +196,7 @@ const DatasetImportContextProvider = ({ children }: { children: React.ReactNode
      chunkSize: vectorModel?.defaultToken ? vectorModel?.defaultToken * 2 : 1024,
      showChunkInput: false,
      showPromptInput: false,
-      charsPointsPrice: agentModel.charsPointsPrice,
+      charsPointsPrice: agentModel.charsPointsPrice || 0,
      priceTip: t('dataset:import.Auto mode Estimated Price Tips', {
        price: agentModel.charsPointsPrice
      }),
@@ -211,7 +211,7 @@ const DatasetImportContextProvider = ({ children }: { children: React.ReactNode
      chunkSize: embeddingChunkSize,
      showChunkInput: true,
      showPromptInput: false,
-      charsPointsPrice: vectorModel.charsPointsPrice,
+      charsPointsPrice: vectorModel.charsPointsPrice || 0,
      priceTip: t('dataset:import.Embedding Estimated Price Tips', {
        price: vectorModel.charsPointsPrice
      }),
@@ -226,7 +226,7 @@ const DatasetImportContextProvider = ({ children }: { children: React.ReactNode
      chunkSize: qaChunkSize,
      showChunkInput: true,
      showPromptInput: true,
-      charsPointsPrice: agentModel.charsPointsPrice,
+      charsPointsPrice: agentModel.charsPointsPrice || 0,
      priceTip: t('dataset:import.Auto mode Estimated Price Tips', {
        price: agentModel.charsPointsPrice
      }),
--- a/projects/app/src/service/events/generateQA.ts
+++ b/projects/app/src/service/events/generateQA.ts
@@ -12,7 +12,10 @@ import { getLLMModel } from '@fastgpt/service/core/ai/model';
 import { checkTeamAiPointsAndLock } from './utils';
 import { checkInvalidChunkAndLock } from '@fastgpt/service/core/dataset/training/utils';
 import { addMinutes } from 'date-fns';
-import { countGptMessagesTokens } from '@fastgpt/service/common/string/tiktoken/index';
+import {
+  countGptMessagesTokens,
+  countPromptTokens
+} from '@fastgpt/service/common/string/tiktoken/index';
 import { pushDataListToTrainingQueueByCollectionId } from '@fastgpt/service/core/dataset/training/controller';
 import { loadRequestMessages } from '@fastgpt/service/core/chat/utils';
 import { llmCompletionsBodyFormat } from '@fastgpt/service/core/ai/utils';
@@ -153,7 +156,8 @@ ${replaceVariable(Prompt_AgentQA.fixedText, { text })}`;
      pushQAUsage({
        teamId: data.teamId,
        tmbId: data.tmbId,
-        tokens: await countGptMessagesTokens(messages),
+        inputTokens: await countGptMessagesTokens(messages),
+        outputTokens: await countPromptTokens(answer),
        billId: data.billId,
        model: modelData.model
      });
--- a/projects/app/src/service/events/generateVector.ts
+++ b/projects/app/src/service/events/generateVector.ts
@@ -111,7 +111,7 @@ export async function generateVector(): Promise<any> {
    pushGenerateVectorUsage({
      teamId: data.teamId,
      tmbId: data.tmbId,
-      tokens,
+      inputTokens: tokens,
      model: data.model,
      billId: data.billId
    });
--- a/projects/app/src/service/support/wallet/usage/push.ts
+++ b/projects/app/src/service/support/wallet/usage/push.ts
@@ -37,7 +37,8 @@ export const pushChatUsage = ({
      moduleName: item.moduleName,
      amount: item.totalPoints || 0,
      model: item.model,
-      tokens: item.tokens
+      inputTokens: item.inputTokens,
+      outputTokens: item.outputTokens
    }))
  });
  addLog.info(`finish completions`, {
@@ -52,20 +53,23 @@ export const pushQAUsage = async ({
  teamId,
  tmbId,
  model,
-  tokens,
+  inputTokens,
+  outputTokens,
  billId
 }: {
  teamId: string;
  tmbId: string;
  model: string;
-  tokens: number;
+  inputTokens: number;
+  outputTokens: number;
  billId: string;
 }) => {
  // 计算价格
  const { totalPoints } = formatModelChars2Points({
    model,
    modelType: ModelTypeEnum.llm,
-    tokens
+    inputTokens,
+    outputTokens
  });

  concatUsage({
@@ -73,7 +77,8 @@ export const pushQAUsage = async ({
    teamId,
    tmbId,
    totalPoints,
-    tokens,
+    inputTokens,
+    outputTokens,
    listIndex: 1
  });

@@ -84,30 +89,32 @@ export const pushGenerateVectorUsage = ({
  billId,
  teamId,
  tmbId,
-  tokens,
+  inputTokens,
  model,
  source = UsageSourceEnum.fastgpt,
  extensionModel,
-  extensionTokens
+  extensionInputTokens,
+  extensionOutputTokens
 }: {
  billId?: string;
  teamId: string;
  tmbId: string;
-  tokens: number;
+  inputTokens: number;
  model: string;
  source?: UsageSourceEnum;

  extensionModel?: string;
-  extensionTokens?: number;
+  extensionInputTokens?: number;
+  extensionOutputTokens?: number;
 }) => {
  const { totalPoints: totalVector, modelName: vectorModelName } = formatModelChars2Points({
    modelType: ModelTypeEnum.vector,
    model,
-    tokens
+    inputTokens
  });

  const { extensionTotalPoints, extensionModelName } = (() => {
-    if (!extensionModel || !extensionTokens)
+    if (!extensionModel || !extensionInputTokens)
      return {
        extensionTotalPoints: 0,
        extensionModelName: ''
@@ -115,7 +122,8 @@ export const pushGenerateVectorUsage = ({
    const { totalPoints, modelName } = formatModelChars2Points({
      modelType: ModelTypeEnum.llm,
      model: extensionModel,
-      tokens: extensionTokens
+      inputTokens: extensionInputTokens,
+      outputTokens: extensionOutputTokens
    });
    return {
      extensionTotalPoints: totalPoints,
@@ -132,7 +140,7 @@ export const pushGenerateVectorUsage = ({
      tmbId,
      totalPoints,
      billId,
-      tokens,
+      inputTokens,
      listIndex: 0
    });
  } else {
@@ -147,7 +155,7 @@ export const pushGenerateVectorUsage = ({
          moduleName: 'support.wallet.moduleName.index',
          amount: totalVector,
          model: vectorModelName,
-          tokens
+          inputTokens
        },
        ...(extensionModel !== undefined
          ? [
@@ -155,7 +163,8 @@ export const pushGenerateVectorUsage = ({
                moduleName: 'core.module.template.Query extension',
                amount: extensionTotalPoints,
                model: extensionModelName,
-                tokens: extensionTokens
+                inputTokens: extensionInputTokens,
+                outputTokens: extensionOutputTokens
              }
            ]
          : [])
@@ -166,17 +175,20 @@ export const pushGenerateVectorUsage = ({
 };

 export const pushQuestionGuideUsage = ({
-  tokens,
+  inputTokens,
+  outputTokens,
  teamId,
  tmbId
 }: {
-  tokens: number;
+  inputTokens: number;
+  outputTokens: number;
  teamId: string;
  tmbId: string;
 }) => {
  const qgModel = global.llmModels[0];
  const { totalPoints, modelName } = formatModelChars2Points({
-    tokens,
+    inputTokens,
+    outputTokens,
    model: qgModel.model,
    modelType: ModelTypeEnum.llm
  });
@@ -192,7 +204,8 @@ export const pushQuestionGuideUsage = ({
        moduleName: 'core.app.Question Guide',
        amount: totalPoints,
        model: modelName,
-        tokens
+        inputTokens,
+        outputTokens
      }
    ]
  });
@@ -215,7 +228,7 @@ export function pushAudioSpeechUsage({
 }) {
  const { totalPoints, modelName } = formatModelChars2Points({
    model,
-    tokens: charsLength,
+    inputTokens: charsLength,
    modelType: ModelTypeEnum.audioSpeech
  });

@@ -251,7 +264,7 @@ export function pushWhisperUsage({

  const { totalPoints, modelName } = formatModelChars2Points({
    model: whisperModel.model,
-    tokens: duration,
+    inputTokens: duration,
    modelType: ModelTypeEnum.whisper,
    multiple: 60
  });