[NeurIPS 2026] DataFlex: A Unified Benchmark and Evaluation Platform for Data-Centric Training of Large Language Models
DataFlex is a data-centric training framework that enhances model performance by either selecting the most influential samples, optimizing their weights, or adjusting their mixing ratios.
A Data-Centric RL and On-Policy Distillation Framework for LLM Post-Training
(ACL 2026 Main) LLMSurgeon recovers the pretraining data mixture of any LLM from only its generated text — no weights, no training data. A calibrated domain classifier plus label-shift correction de-blurs biased predictions. Ships with LLMScan, a benchmark on 8 open-source LLMs.
LLM pretraining data mixture research insights
Exact token-weighted cluster totals and reproducible calibration for NVIDIA Nemotron-ClimbMix
Unofficial reproduction of Nemotron-CLIMB (CLustering-based Iterative Data Mixture Bootstrapping, NeurIPS 2025, arXiv:2504.13161) on Ascend NPUs — 5 rounds, ≈3,900 NPU-hours; learned mixtures beat every fixed-ratio baseline (+0.014–0.043 STEM); the selection mechanism is the weak link — fixed and validated live.
To associate your repository with the data-mixture topic, visit your repo's landing page and select "manage topics."