Event ·Multi-source

Nemotron 3.5 Lightning微调模型，30B形式化证明胜过50倍

Reasonable 团队用 4B token 的合成 Verus 数据微调了 NVIDIA 最新开源模型 Nemotron 3.5 Lightning（30B）。微调后的模型在单次尝试通过率上超越了一个规模约 50 倍于它的模型，pass@3 则几乎持平。它的 token 生成速度也快于测试过的所有同尺寸开源模型。配套博客还分析了各模型在形式化证明中的失败方式、作弊行为，以及微调带来的影响。

Current read

Reasonable微调了NVIDIA的30B开源模型，形式化证明通过率胜过50倍大的模型，生成还更快。博客还拆解了模型怎么作弊。

11 sources58° AI heatLast updated 2026/8/12 21:47:41

Evidence

11 sources

Primary sourceNVIDIA AI

2026/8/12 21:47:41View source

Related source 1Decoder

2026/8/11 15:07:44View source

Related source 2ollama

2026/8/11 16:02:21View source

Related source 3Harrison Chase

2026/8/11 17:23:48View source

Related source 4IT之家

2026/8/12 01:49:32View source

Related source 5AI Will

2026/8/12 06:41:07View source

Related source 6marktechpost

2026/8/12 06:59:40View source

Related source 7OpenRouter

2026/8/11 13:17:30View source

Related source 8Fireworks AI

2026/8/11 14:49:20View source

Related source 9Aravind Srinivas

2026/8/11 19:00:43View source

Related source 10Julien Chaumond

2026/8/12 16:16:12View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story