From the source
Ant Ling has open-sourced the completed Base Models of Ling-3.0-tiny and Ling-3.0-flash, along with pre-training and mid-training checkpoints, after a community member asked for a Base Model.
- The company says the Base Models are the point, and the earlier weights make them more useful by opening a training path rather than only its endpoint.
- Ling-3.0-tiny-base has 7.9B total parameters and 1.3B activated parameters, while Ling-3.0-flash-base has 124B total parameters and 5.1B activated parameters.
- For both models Ant Ling released a pre-training checkpoint, a mid-training checkpoint, and the completed Base Model, and after mid-training Ling-3.0 uses WSM, a checkpoint-merging approach that supports continued pre-training, dynamic data expansion, and offline exploration of different training trajectories.
- Ant Ling says the models are not instruction-aligned chat products and should not be deployed directly to end users without task-specific post-training, evaluation, and validation.
