Announcement
Nemotron 3.5 Super VL fine-tuning is claimed to raise performance from 43% to 73% on the hardest benchmark
ReasonableIO says it tested whether the model could write challenging Verus proofs for TLA+ specifications.
TLDR
ReasonableIO says Nvidia’s Nemotron 3.5 Super VL could write challenging Verus proofs for TLA+ specifications in its tests. It reports that its fine-tuning raised performance from 43% to 73% on the hardest benchmark, and links to a report on novel datasets, fine-tuning and the gap to real-world software verification.
Combined views
1.2K
3 Sources, first seen ago