OpenDCAI/DataFlex
A dynamic training framework for LLMs that performs data selection, domain mixture optimization, and example reweighting during the training process.
Collecting fresh signals — velocity needs a few days of history.
collecting data…
star history
DataFlex is an open-source framework for enhancing LLM training through intelligent data curation. It dynamically adjusts training data during the training loop by selecting samples, optimizing domain mixtures, and reweighting examples based on training signals. Built on top of LLaMA-Factory, it integrates seamlessly into existing LLM fine-tuning workflows and supports distributed training with DeepSpeed ZeRO-3.