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

NVIDIA Releases Nemotron-Labs-3-Puzzle-75B-A9B: A Compressed Hybrid MoE LLM Delivering 2.03x Server Throughput at Matched User Throughput

精选理由

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。

AI 摘要

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。

AI 翻译 · 中文

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。

marktechpostNVIDIA has released Nemotron-Labs-3-Puzzle-75B-A9B, a compressed variant of Nemotron-3-Super. Iterative Puzzle alternates hardware-aware structural compression with short knowledge distillation recovery phases. The model