- [2025-12-23] 🎉 We’re excited to announce the first Data-Centric Training System DataFlex, is now released! Stay tuned for future updates.
DataFlex is an advanced dynamic training framework built on top of LLaMA-Factory.
It intelligently schedules data during training, supporting dynamic sample selection, domain ratio adjustment, and dynamic weighting, aiming to improve both training efficiency and final model performance.
DataFlex integrates seamlessly with LlamaFactory, offering researchers and developers more flexible and powerful training control, for goals and design philosophy, please refer to Dataflex-Doc.
- Dynamic Select Trainer: Dynamically selects training samples according to a given strategy (e.g., focus on “hard” samples). The data selection algorithms are summarized as follows:
| Method | Category | Requires Model-in-the-Loop? |
|---|---|---|
| LESS | Gradient-Based | ✅ Yes |
| NICE | Gradient-Based | ✅ Yes |
| Loss | Loss-Based | ✅ Yes |
| Delta Loss | Loss-Based | ✅ Yes |
| NEAR | Data Distribution-Based | ❌ No |
| TSDS | Data Distribution-Based | ❌ No |
| Static | No Selection | ❌ No |
| Random | Random Sampling | ❌ No |
- Dynamic Mix Trainer: Dynamically adjusts the ratio of data from different domains during training. The data mixture algorithms are summarized as follows:
| Method | Category | Requires Model-in-the-Loop? |
|---|---|---|
| DOREMI | Offline Mixture | ✅ Yes |
| ODM | Online Mixture | ✅ Yes |
- Dynamic Weight Trainer: Dynamically adjusts sample weights during backpropagation to emphasize data preferred by the model. The data reweighting algorithms are summarized as follows:
| Method | Category | Requires Model-in-the-Loop? |
|---|---|---|
| Loss Reweighting | Loss-Based | ✅ Yes |
- Full compatibility with LlamaFactory, drop-in replacement.
Please use the following commands for environment setup and installation👇
git clone https://github.com/OpenDCAI/DataFlex.git
cd DataFlex
pip install -e .
pip install llamafactory==0.9.3The launch command is similar to LLaMA-Factory. Below is an example using LESS :
FORCE_TORCHRUN=1 DISABLE_VERSION_CHECK=1 dataflex-cli train examples/train_lora/selectors/less.yamlUnlike vanilla LlamaFactory, your .yaml config file must also include DataFlex-specific parameters, for details please refer to DataFlex-Doc.
Using DataFlex can improve performance over the default LLaMA-Factory training.
We use a subset of Open-Hermes-2.5 as the training dataset. The data selection algorithms and data reweighting algorithm outperform the random selector baseline on the MMLU benchmark subset relevant to the training dataset. For the Less and Nice algorithm, we set the validation set as the MMLU-Validation-Set, using a GPT-5-generated trajectory.
We use a subset of SlimPajama-627B for data mixture。The data mixture algorithm also outperforms baselines (default data mixture) on the MMLU benchmark.
| Dataset | Baseline | DoReMi | ODM |
|---|---|---|---|
| MMLU | 25.27 | 25.84 | 26.04 |
We thank LLaMA-Factory for offering an efficient and user-friendly framework for large model fine-tuning, which greatly facilitated rapid iteration in our training and experimentation workflows.
Our gratitude extends to all contributors in the open-source community—their efforts collectively drive the development of DataFlex.
We welcome contributions of new trainers and selectors! Please ensure code formatting is consistent with the existing style before submitting a PR.
We also welcome you to join the DataFlex and DataFlow open-source community to ask questions, share ideas, and collaborate with other developers!
• 📮 GitHub Issues: Report bugs or suggest features
• 🔧 GitHub Pull Requests: Contribute code improvements
• 💬 Join our community groups to connect with us and other contributors!
