Event ·Official first-hand

压缩混合MoE模型Nemotron-Labs-3-Puzzle-75B-A9B，吞吐量提升2.03倍

NVIDIA发布Nemotron-Labs-3-Puzzle-75B-A9B，这是Nemotron-3-Super的压缩变体。通过Iterative Puzzle方法，总参数从120.7B降至75.3B，活跃参数从12.8B降至9.3B。在单台8xB200节点上，它实现了2.03倍于Super的总吞吐量，每用户100 tok/s。在H100上，1M-token并发从1提升到8。

Current read

NVIDIA把这个MoE模型压到75B，吞吐翻倍，用户延迟不变，适合降本增效。

11 sources73° AI heatLast updated 2026/7/9 08:47:37

Evidence

11 sources

Primary sourcemarktechpost

2026/7/9 08:47:37View source

Related source 1elvis

2026/7/7 17:20:02View source

Related source 2NVIDIA AI

2026/7/8 01:37:24View source

Related source 3LangChain

2026/7/8 16:13:53View source

Related source 4Decoder

2026/7/8 19:01:44View source

Related source 5IT之家

2026/7/8 22:51:25View source

Related source 6Aravind Srinivas

2026/7/9 14:38:08View source

Related source 7arXiv cs.LG

2026/7/9 17:41:49View source

Related source 8AWS Machine Learning Blog

2026/7/10 15:35:05View source

Related source 9Harrison Chase

2026/7/10 16:39:30View source

Related source 10Clement Delangue

2026/7/10 19:57:28View 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