ByteDance's Seed team released a diffusion language model with an inference speed of 2146 tokens per second.

2025-07-31 20:37:11

On July 31st, the ByteDance Seed team released Seed Diffusion Preview, an experimental diffusion language model. The team's goal, according to the team, is to systematically validate the feasibility of the discrete diffusion technology approach as a foundational framework for next-generation language models, using structured code generation as an experimental field. Experimental results show that Seed Diffusion Preview achieves an inference speed of 2146 tokens/s, a 5.4x increase compared to an autoregressive model of similar scale.

ByteDance

Email Subscription

Newsletters and emails are now available! Delivered on time, every weekday, to keep you up to date with North American business news.

Weekly Highlights

                                    Beijing, the "capital of open source", accelerates the construction of domestic open source big models and achieves another breakthrough
                            2025-07-25

                                    Novo Nordisk (NVO.N): The downward revision of its fiscal year sales forecast is related to declining sales of its weight loss drug Wegovy in the United States.
                            2025-07-29

                                    Toyota Motor Corp. is expected to decide as early as Wednesday evening whether to resume operations at its Japanese factories.
                            2025-07-30

                                    The world's first active safety AI battery cell goes into mass production
                            2025-07-27

                                    CATL's losses widened in the afternoon, with A shares falling more than 4% to below 280 yuan and H shares falling more than 7%.
                            2025-07-30