Nemotron-3-Embed
model Your tags
Your notes
NVIDIA's open embedding collection (8B and 1B, BF16 and NVFP4): the 8B ranks #1 overall on RTEB (78.46 avg NDCG@10 at release). Bidirectional-attention Transformer encoders, 32K context, 34 languages, with an NVFP4 Blackwell 4-bit path retaining ~99.5% of BF16 accuracy at ~2× throughput.
Model Details
Parameters 8B
Context window 32,768