Currently, Dataflow integrates five text data generators, covering various formats such as pretraining document data, SFT-format data, and multi-turn dialogues.

Name	Applicable Type	Description	Repository or Paper
PretrainGenerator	Pretrain	Synthesize phi-4 question and answer data pairs using pre trained document data, and retell the document in QA format	Paper
SFTGeneratorSeed	SFT	Synthesize SFT format QA data pairs based on seed documents and return original information	-
CondorGenerator	SFT	Two-stage synthesis of SFT-format data from scratch based on preset knowledge tree labels (recommend increasing label variety if generating more than 5000 samples)	paper
PromptedGenerator	-	Generate data based on user-defined prompts	-
ConsistentChatGenerator	Multi-turn Dialogue	Two-stage synthesis of multi-turn dialogue data from scratch based on preset topics and human intents (recommend increasing label variety if generating more than 9000 samples)	paper