NVIDIA released the Nemotron-Labs-Diffusion model family (3B to 14B) that supports both AR and diffusion decoding with novel self-speculation, achieving significant speedups (up to 4x) over standard AR and Eagle3 methods across hardware platforms.
Model Overview Nemotron-Labs-Diffusion is a tri-mode language model that supports both AR decoding and diffusion-based parallel decoding by simply switching the attention pattern of the same model during inference. The synergy between these two modes enables a third mode, called self-speculation: the same model performs diffusion-based parallel drafting and AR verification with shared KV cache, achieving high acceptance lengths and decoding efficiency. The seamless mode switching by simply changing attention patterns enables high efficiency at different concurrency levels in varying deployment scenarios with one single model. https://preview.redd.it/mwyq7b7hx42h1.png?width=3915&format=png&auto=webp&s=744bd87267338a6236269a8d915b185cff8a82d2 # Highlights * SOTA 3B, 8B, 14B dense LM family (base, instruct, and vision-language variants) supporting AR, diffusion, and self-speculation with the focus on decode efficiency. * Generation moved from a memory-bound regime toward a compute-bound regime. Model weights are loaded once and reused to compute multiple tokens during generation. * Self-speculation uses diffusion for drafting and AR for verification, providing a stronger alternative to MTP approaches: * 3x higher acceptance length and 2.2x speed-up vs. Qwen3-8B-Eagle3 in SGLang. * 5.9× tokens per forward over Qwen3-8B (no MTP) with the same accuracy. * Real-device speed-up across platforms: * DGX Spark (8B, concurrency 1): 2.7x faster with 112 tok/sec vs. 41.8 tok/sec AR using w4a16. * GB200 (8B, concurrency 1): 3.3x faster with 850 tok/sec vs. 253 tok/sec AR and 360 tok/sec Eagle3. Custom CUDA kernels boost to 1015 tok/sec (4x). * Diffusion speedup-of-light analysis shows that throughput can be further doubled (vs. current best) for a single user with better sampling - future research. [https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-VLM-8B](https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-VLM-8B) [https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-14B-Base](https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-14B-Base) [https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-14B](https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-14B) [https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-8B-Base](https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-8B-Base) [https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-8B](https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-8B) [https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-3B-Base](https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-3B-Base) [https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-3B](https://huggingface.co/nvidia/Nemotron-Labs-Diffusion-3B) #
NVIDIA releases Nemotron-Labs-Diffusion, a family of tri-mode language models (3B, 8B, 14B) supporting AR, diffusion, and self-speculation decoding, achieving 2.7x-4x speed-ups over standard AR decoding.
NVIDIA releases Nemotron-Labs-Diffusion, the first tri-mode language model family (3B/8B/14B) that switches between autoregressive, diffusion, and self-speculation decoding by changing the attention pattern, achieving up to 4× higher real throughput.
The paper introduces Nemotron-Labs-Diffusion, a tri-mode language model that unifies autoregressive, diffusion, and self-speculation decoding, achieving superior throughput and efficiency compared to existing models.
NVIDIA released Nemotron-TwoTower-30B-A3B-Base-BF16, a diffusion-based language model that uses block-wise autoregressive diffusion to generate text by iterative denoising of token blocks, achieving 2.42× the generation throughput of the autoregressive baseline while retaining 98.7% of benchmark quality.
NVIDIA introduces Nemotron-Labs Diffusion, a family of diffusion language models that generate text in parallel and iteratively refine it, offering faster generation and the ability to revise previous tokens.