NVIDIA 开源了三个嵌入模型，8B 的 RTEB 第一，1B 的量化版速度翻倍还能保持 99% 精度，适合做检索和 RAG。
NVIDIA 于 2026 年 7 月 15-16 日发布 Nemotron 3 Embed 系列，包含三个开源检查点：8B-BF16、1B-BF16 和 1B-NVFP4。8B 模型在 RTEB 基准上以 78.46 平均 NDCG@10 排名第一。1B 模型通过 ModelOpt NAS 剪枝和 COS+MSE 蒸馏从 8B 教师模型获得。NVFP4 量化版本在 Blackwell 上吞吐量提升 2 倍，同时保留超过 99% 的 BF16 检索精度。所有检查点支持 32,768 token 输入，基于 OpenMDW-1.1 协议。
NVIDIA AI Releases Nemotron 3 Embed: An Open Embedding Collection Whose 8B Checkpoint Ranks #1 on RTEB
NVIDIA released Nemotron 3 Embed on July 15 and 16, 2026. The collection has three open checkpoints: Nemotron-3-Embed-8B-BF16, Nemotron-3-Embed-1B-BF16, and Nemotron-3-Embed-1B-NVFP4. The 8B ranks #1 on RTEB at 78.46 average NDCG@10. The 1B came from ModelOpt NAS pruning plus COS+MSE distillation from the 8B teacher. NVFP4 retains 99%+ of BF16 retrieval accuracy at up to 2x Blackwell throughput. All three run 32,768-token inputs under OpenMDW-1.1. The post NVIDIA AI Releases Nemotron 3 Embed: An Open Embedding Collection Whose 8B Checkpoint Ranks #1 on RTEB appeared first on MarkTechPost .