Intern-S2-Preview-OPD is a post-trained version of
Intern-S2-Preview, an efficient 35B scientific multimodal foundation model.
The model is post-trained using the method introduced in
SimpleOPD: Simple Tokenizer-Agnostic On-Policy Distillation of Long-Context Reasoning.
This post-training process substantially improves the model's reasoning performance, particularly on challenging proof and mathematical reasoning benchmarks.
The values in parentheses indicate absolute improvements over the base model.
Intern-S2-Preview-OPD uses the same model architecture and inference interface as Intern-S2-Preview. Please refer to the
Intern-S2-Preview model card
for deployment instructions and recommended inference settings.
This model is released under the
Apache License 2.0.
