Description
Preparing LLM training data means scripts for every cleaning, filtering, synthesis and rewriting step that are hard to reuse. DataFlow makes data preparation easy with LLM-based operators and pipelines for cleaning, synthesis and quality evaluation.
It ships many operators and preset pipelines with vLLM and SGLang backends and a Gradio UI, published at SIGMOD 2027.
LLM operators:Clean, synthesize and evaluate.
Pipelines:Ready to use.
Backends:vLLM and SGLang.
UI:Gradio.
It ships many operators and preset pipelines with vLLM and SGLang backends and a Gradio UI, published at SIGMOD 2027.
Features
LLM operators:Clean, synthesize and evaluate.
Pipelines:Ready to use.
Backends:vLLM and SGLang.
UI:Gradio.
