
Ant Baoling Open-sources Ling-3.0 tiny / flash Base Models, Releases Key Checkpoints During Training Process
Ant Baoling open-sources Ling-3.0-tiny-base and Ling-3.0-flash-base models, simultaneously releasing pre-training and mid-training weights, totaling 6 checkpoints, to assist developer research.
Ant Baoling has officially open-sourced the Ling-3.0-tiny-base and Ling-3.0-flash-base models. This release includes not only the final weights but also key checkpoints from the pre-training and mid-training phases.
This open-source release provides a total of 6 checkpoints, breaking the previous limitation of only releasing final models, and providing researchers with a complete data chain to observe the evolution of model capabilities.
Releasing weights during the training process helps the academic community deeply analyze the emergence mechanisms of large models during the pre-training and mid-training stages, lowering the threshold for reproduction and improvement.
With the release of more intermediate weight data, the open-source community will be able to conduct model distillation or fine-tuning experiments more effectively, promoting transparency and collaborative development of foundation model technology.
This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.