NVIDIA新嵌入模型Nemotron-3-Embed在检索任务上超越Qwen-3-600m，开放权重加NVFP4，适合做记忆系统。
Mem0在其记忆检索管线中评估了NVIDIA的Nemotron-3-Embed嵌入模型。在LongMemEval基准上，该模型的Recall@10从Qwen-3-600m的78.71提升至80.38。Nemotron-3-Embed具有开放权重，支持Blackwell上的NVFP4高效推理，且吞吐量高，适合频繁写入的系统。Mem0还详细介绍了其嵌入使用方式及评估过程。
We evaluated Nemotron-3-Embed from @NVIDIAAI for our @mem0ai memory retrieval pipeline Tested on lo...
We evaluated Nemotron-3-Embed from @NVIDIAAI for our @mem0ai memory retrieval pipeline Tested on longmemeval: retrieval @10 improved from 78.71 to 80.38 over qwen-3-600m A few things made it a good fit for how we operate: open weights, NVFP4 support on Blackwell for efficient inference, and strong throughput, which matters for a system where writes happen constantly. A breakdown on how Mem0 uses embeddings and Nemotron-3-embed evaluation below👇 💬 2 🔄 1 ❤️ 4 👀 278 📊 4 ⚡