perf: buffer;fix: back up split (#4913)

* perf: buffer

* fix: back up split

* fix: app limit

* doc

packages/service/core/dataset/collection/controller.ts

@@ -77,7 +77,10 @@ export const createCollectionAndInsertData = async ({
   const chunkSplitter = computeChunkSplitter(createCollectionParams);
   const paragraphChunkDeep = computeParagraphChunkDeep(createCollectionParams);
 
-  if (trainingType === DatasetCollectionDataProcessModeEnum.qa) {
+  if (
+    trainingType === DatasetCollectionDataProcessModeEnum.qa ||
+    trainingType === DatasetCollectionDataProcessModeEnum.backup
+  ) {
     delete createCollectionParams.chunkTriggerType;
     delete createCollectionParams.chunkTriggerMinSize;
     delete createCollectionParams.dataEnhanceCollectionName;

packages/service/core/dataset/read.ts

@@ -218,6 +218,10 @@ export const rawText2Chunks = ({
     };
   };
 
+  if (backupParse) {
+    return parseDatasetBackup2Chunks(rawText).chunks;
+  }
+
   // Chunk condition
   // 1. 选择最大值条件，只有超过了最大值(默认为模型的最大值*0.7），才会触发分块
   if (chunkTriggerType === ChunkTriggerConfigTypeEnum.maxSize) {
@@ -240,10 +244,6 @@ export const rawText2Chunks = ({
     }
   }
 
-  if (backupParse) {
-    return parseDatasetBackup2Chunks(rawText).chunks;
-  }
-
   const { chunks } = splitText2Chunks({
     text: rawText,
     chunkSize,