论文58
NVIDIA 发布 Nemotron-Labs-Sigma 开源语言扩散模型
原标题:Language generation beyond one token at a time. 💬
NVIDIA 发布 Nemotron-Labs-Sigma 系列开放模型权重,共 14 个 checkpoint,规模为 3B 和 8B。该模型把连续扩散用于语言生成,可并行生成 token 块,为探索数学和编码问题的多种解法提供更快的生成路径。论文见 arxiv.org/abs/2610.02665,模型已在 Hugging Face 开放。
中文译文尚未完成,先为你展示原文。
Language generation beyond one token at a time. 💬
NVIDIA Nemotron-Labs-Sigma generates blocks of tokens in parallel, opening the door to faster generation for exploring multiple solutions to math and coding problems.
Read the paper: http://arxiv.org/abs/2610.02665
Try the open models: http://huggingface.co/collections/nvidia/nemotron-labs-sigma
📣 NVIDIA is releasing a collection of open model weights: Nemotron-Labs-Sigma — 14 checkpoints at 3B/8B. Language models have stayed discrete. We bring continuous diffusion to language as in vision. Smooth ODE trajectories (16-D) unlock distillation and steering: 8 TPFs for math, 4 TPFs for code, with accuracy and diversity preserved. 🧵 1/查看原帖
