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>

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
       });