NVIDIA 发布 Nemotron 3 Embed 系列，8B 版本在 RTEB 基准上排名第一

嵌入模型决定了代理能看到哪些段落。NVIDIA 发布了 Nemotron 3 Embed 模型：https://huggingface.co/collections/nvidia/nemotron-3-embed 用于在该层工作。它针对生产规模的 RAG、智能检索、代码检索和代理记忆。

模型集合包括三个开放检查点。Nemotron-3-Embed-8B-BF16 是优先考虑准确性的选项。Nemotron-3-Embed-1B-BF16 将相同设计压缩到更小的规模。Nemotron-3-Embed-1B-NVFP4 是经过 Blackwell 优化的 4 位路径。

这三者都是使用双向注意力掩码训练的 Transformer 编码器。最终嵌入来自对 token 级表示的平均池化。所有检查点的最大序列长度为 32,768 个 token。

每个模型在 34 种语言中进行了评估。三者均遵循 OpenMDW 许可证协议，第 1.1 版 (OpenMDW-1.1)。值得注意的是，基础模型为 Mistral 模型。8B 版本基于 Ministral-3-8B-Instruct-2512 构建。两个 1B 变体使用 Ministral-3-3B-Instruct-2512。

Nemotron-3-Embed-8B-BF16 在 RTEB 上总体排名第 1：https://mteb-leaderboard.hf.space/benchmark/RTEB(beta) （截止至 2026 年 7 月 17 日），即检索嵌入基准。评估覆盖其 16 个公开任务。下列每个数值均为 NDCG@10 的平均值，模型序列长度为 4096。

值得注意的两个差距。1B 在 RTEB 上比上一代基线 llama-nemotron-embed-vl-1b-v2 提升 10.4 分。另一方面，NVFP4 与其 BF16 父模型相比损失 0.38 RTEB 分，相当于保持率 99.5%。

这些 1B 分数来自压缩流程，而非更小规模的训练。父模型为 nemotron-3-embed-3b，经两轮迭代修剪和蒸馏得到。

首先，3B 父模型使用 NVIDIA ModelOpt mcore_minitron 神经架构搜索 (NAS) 修剪至 2B。搜索涵盖隐藏层宽度、FFN 大小、注意力头数和深度，然后从前 10 名 Pareto 前沿中选择最佳候选。50k 的域内校准语料对这些候选进行评分。

接下来，从微调后的 8B 嵌入教师模型蒸馏得到 2B 模型。蒸馏结合了余弦距离损失 (COS) 和均方误差损失 (MSE)。数据混合为多语言和域内数据。最后，同样的流程重复，生成 1.14B 检查点。

压缩然后继续进入服务格式。量化只影响线性层的权重和激活，目标数据类型为 NVFP4。研究团队使用了 nvidia-modelopt v0.45.0。随后进行了量化感知蒸馏（QAD），主要用于恢复长输入的准确性。

校准使用了 512 个样本：256 个查询和 256 个来自 abisee/cnn_dailymail 的段落。QAD 训练使用了 20k 个样本。

研究团队报告称，Blackwell 上的 NVFP4 提供的吞吐量比 BF16 高出最多 2 倍。它保留了 BF16 检索准确率的 99%。NVFP4 卡还支持动态嵌入大小。你可以将 2048 维向量从开头切到 1024 或 512 维度，之后进行重新归一化。

在操作代码之前，先观察路径运行。它会展示前缀添加、双向编码、平均池化、L2 归一化和点积评分。评分来自每个卡发布的预期输出。

正如该演示所示，检查点不共享运行时路径。

除了检查点外，NVIDIA 研究团队发布了针对 1B 模型的优化 NIM 微服务。基于 Rust 的 NIM 在 GB200 和 RTX PRO 6000 上与 vLLM 检查点匹配或表现更佳。NVIDIA 测试了输入序列长度为 256 和 1024。另有 NVIDIA NeMo AutoModel 配方涵盖微调和蒸馏。

考虑到这些路径，前缀先行。查询使用 query:，文档使用 passage:. 嵌入向量经过 L2 归一化，因此点积等于余弦相似度。

encode_query 和 encode_document 会读取保存的提示。因此你无需手动添加前缀。对于服务端，/v2/embed 会根据 input_type 应用前缀：

需要与我们合作推广你的 GitHub 仓库、Hugging Face 页面、产品发布或网络研讨会等？请与我们联系：https://forms.gle/wbash1wF6efRj8G58

Asif Razzaq 是 Marktechpost Media Inc. 的首席执行官。作为一位有远见的企业家和工程师，Asif 致力于利用人工智能的潜力造福社会。他最近的努力是推出人工智能媒体平台 Marktechpost，该平台因其对机器学习和深度学习新闻的深入报道而脱颖而出，既技术上可靠，又易于广大受众理解。该平台每月浏览量超过 200 万次，显示了其在观众中的受欢迎程度。

Embedding models decide which passages an agent ever sees. NVIDIA released Nemotron 3 Embed model：https://huggingface.co/collections/nvidia/nemotron-3-embed to work on that layer. It targets production-scale RAG, agentic retrieval, code retrieval, and agent memory.

The model collection includes three open checkpoints. Nemotron-3-Embed-8B-BF16 is the accuracy-first option. Nemotron-3-Embed-1B-BF16 carries the same design into a smaller footprint. Nemotron-3-Embed-1B-NVFP4 is the Blackwell-optimized 4-bit path.

All three are transformer encoders trained with bidirectional attention masking . The final embedding comes from average pooling over token-level representations. Maximum sequence length is 32,768 tokens on every checkpoint.

Each model was evaluated across 34 languages. All three carry the OpenMDW License Agreement, version 1.1 (OpenMDW-1.1) . Notably, the bases are Mistral models. The 8B is built with Ministral-3-8B-Instruct-2512 . Both 1B variants use Ministral-3-3B-Instruct-2512 .

Nemotron-3-Embed-8B-BF16 ranks #1 overall on RTEB：https://mteb-leaderboard.hf.space/benchmark/RTEB%28beta%29 (as of July 17 2026), the Retrieval Embedding Benchmark. Evaluation covers its 16 public tasks. Every figure below is average NDCG@10, at model sequence length 4096.

Two gaps are worth noting. The 1B gains 10.4 RTEB points over llama-nemotron-embed-vl-1b-v2 , the prior-generation baseline. Separately, NVFP4 costs 0.38 RTEB points against its BF16 parent, or 99.5% retention.

Those 1B scores come from a compression pipeline, not a smaller training run. The parent was nemotron-3-embed-3b , pruned and distilled across two iterative rounds.

First, the 3B parent was pruned to 2B using NVIDIA ModelOpt mcore_minitron Neural Architecture Search (NAS) . The search covers hidden width, FFN size, attention heads, and depth. It then picks the best candidate from the top-10 Pareto front. A 50k in-domain calibration corpus scored those candidates.

Next, the 2B model was distilled from the fine-tuned 8B embedding teacher. Distillation combined cosine distance loss (COS) and mean squared error (MSE) loss. The data blend was multilingual and in-domain. Finally, the same procedure repeated to produce the 1.14B checkpoint.

Compression then continues into the serving format. Quantization hit weights and activations of linear layers only, targeting the NVFP4 data type. The research team used nvidia-modelopt v0.45.0 . Quantization-Aware Distillation (QAD) followed, primarily to recover accuracy on long inputs.

Calibration used 512 samples: 256 queries and 256 passages from abisee/cnn_dailymail . QAD training used 20k samples.

The rsesearch team reports NVFP4 on Blackwell delivers up to 2x higher throughput than BF16 . It retains 99%+ of BF16 retrieval accuracy . The NVFP4 card also documents dynamic embedding sizes. You can slice the 2048-d vector from the start to 1024 or 512 dimensions. Re-normalize afterward.

Before touching code, watch the path run. It animates prefixing, bidirectional encoding, average pooling, L2 normalization, and dot-product scoring. Scores come from each card’s published expected output.

As that walkthrough implies, the checkpoints do not share runtime paths.

Alongside the checkpoints, NVIDIA research team released an optimized NIM microservice for the 1B model. The Rust-based NIM matches or outperforms the vLLM checkpoint on GB200 and RTX PRO 6000. NVIDIA tested input sequence lengths of 256 and 1024. Separately, NVIDIA NeMo AutoModel recipes cover fine-tuning and distillation.

With those paths in mind, prefixes come first. Queries take query: and documents take passage: . Embeddings are L2-normalized, so dot product equals cosine similarity.

encode_query and encode_document read the saved prompts. So you never add prefixes by hand. For serving, /v2/embed applies them from input_type instead:

Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us ：https://forms.gle/wbash1wF6efRj8G58

Asif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences.