Nvidia has introduced Nemotron Cascade 2 30B A3B, a new open-source large language model (LLM). This model is built upon the Nemotron 3 Nano Base but includes significant improvements in post-training.

Performance

Early evaluations suggest that Nemotron Cascade 2 30B A3B offers performance comparable to that of models with 120 billion parameters in benchmarks specific to mathematics and code. Further testing is needed to validate these initial results.

Resources

The model is accessible via Hugging Face. Its architecture and training methodologies are detailed in a research paper available on arXiv.