Session 22: Nemotron-Labs-Diffusion: Unifying AR, Diffusion, and Self-Speculation

[Diffusion LLM]

Hello folks,

Nemotron-Labs-Diffusion is a tri-mode language model (LM) that unifies AR, diffusion, and self-speculation decoding within a single architecture. 

Trained with a joint AR-diffusion objective, Nemotron-Labs-Diffusion can switch modes to sustain high throughput across deployment settings and concurrency levels. Their study shows that 

1. AR and diffusion objectives are complementary: diffusion improves lookahead planning, while AR provides left-to-right linguistic priors. 

2. In self-speculation mode, diffusion drafts while AR verifies, outperforming multi-token prediction (MTP) methods in both acceptance rate and real-device efficiency.

3. A speed-of-light analysis further demonstrates diffusion’s long-term potential, with up to 76.5% more tokens per forward pass than self-speculation under an optimal sampler. 

Scaling to 3B, 8B, and 14B parameters, the Nemotron-Labs-Diffusion family, including base, instruct, and vision-language models, consistently outperforms state-of-the-art open-source AR and diffusion LMs in both accuracy and speed. 

For example, Nemotron-Labs-Diffusion-8B decodes 6× more tokens per forward than Qwen3-8B with comparable accuracy, translating to 4× higher throughput on SPEED-Bench with SGLang on a GB200 GPU.

This Monday, Yonggan Fu from NVIDIA Research will present Nemotron-Labs-Diffusion: Unifying AR, Diffusion, and Self-Speculation. 

Title: Nemotron-Labs-Diffusion: Unifying AR, Diffusion, and Self-Speculation

Meeting Link: click here

Time: June 29 (Monday) 1pm ET / 10am PT / 7pm CET / 10:30pm IST

Paper: Nemotron-Labs-Diffusion: A Tri-Mode Language Model Unifying Autoregressive, Diffusion, and Self-Speculation Decoding | Research 

Prior knowledge: 

Fundamentals of discrete diffusion (video by Sasha Rush)

Abstract:

We introduce Nemotron-Labs-Diffusion, a tri-mode language model (LM) that unifies AR, diffusion, and self-speculation decoding within a single architecture. Trained with a joint AR-diffusion objective, Nemotron-Labs-Diffusion can switch modes to sustain high throughput across deployment settings and concurrency levels. Our study shows that (1) AR and diffusion objectives are complementary: diffusion improves lookahead planning, while AR provides left-to-right linguistic priors. (2) In self-speculation mode, diffusion drafts while AR verifies, outperforming multi-token prediction (MTP) methods in both acceptance rate and real-device efficiency. (3) A speed-of-light analysis further demonstrates diffusion’s long-term potential, with up to 76.5% more tokens per forward pass than self-speculation under an optimal sampler. Scaling to 3B, 8B, and 14B parameters, our Nemotron-Labs-Diffusion family, including base, instruct, and vision-language models, consistently outperforms state-of-the-art open-source AR and diffusion LMs in both accuracy and speed. For example, Nemotron-Labs-Diffusion-8B decodes 5.9×more tokens per forward than Qwen3-8B with better accuracy, translating to 4× higher throughput on SPEED-Bench with SGLang on a GB200 GPU.

Session 22: Nemotron-Labs-Diffusion: Unifying AR, Diffusion, and Self-Speculation Jun 29, 2026, 10:00am – Jun 29, 2026, 11:00am (GMT-07:00) Pacific Time - Los Angeles

Yours truly,

Subham, Justin, Zhihan

Website, Twitter, Discord, YouTube