Event ·Multi-source

CoreWeave RL Rollouts 免重部署加载权重，训练循环提速约 15 倍

CoreWeave 发布 RL Rollouts，把强化学习训练中的新 checkpoint 权重直接加载到在线部署里，不打断在途请求，比传统重部署周期快约 15 倍。NVIDIA 和 You.com 用它对 Nemotron 3.5 Lightning 做后训练以适配网页搜索。结果是 BrowseComp 准确率从 36.97% 提升到 45.45%，工具调用次数减少 30.24%。

Current read

CoreWeave 出了个 RL Rollouts，训练时换 checkpoint 不用重新部署，快了 15 倍，NVIDIA 用它把 BrowseComp 从 37% 拉到 45%。

11 sources55° AI heatLast updated 2026/10/18 16:50:01

Evidence

11 sources

Primary sourceCoreWeave

2026/10/18 16:50:01View source

Related source 1Genk

2026/10/16 22:00:00View source

Related source 2Blognone

2026/10/17 01:16:00View source

Related source 3DAIR.AI

2026/10/17 13:41:02View source

Related source 4NVIDIA AI

2026/10/16 21:46:49View source

Related source 5TestingCatalog

2026/10/17 10:49:40View source

Related source 6IT之家

2026/10/18 08:06:36View source

Related source 7Gorden Sun

2026/10/18 14:13:49View source

Related source 8marktechpost

2026/10/19 03:19:18View source

Related source 9VTV Công nghệ

2026/10/19 06:00:00View source

Related source 10VnExpress Số hóa

2026/10/19 08:38:39View 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