IFM/K2-Horizon-7B-Uno · Hugging Face - 5200tps with no quality loss

IFM has released K2-Horizon-7B-Uno, a diffusion-augmented LLM that achieves speeds of up to 5200 tps with no claimed quality loss. This model utilizes a causal LLM architecture and incorporates a plug-and-play diffusion adapter alongside its autoregressive weights. Further details can be found at https://arxiv.org/abs/2609.04010.

Unlike previous models that often sacrifice quality for speed, K2-Horizon-7B-Uno claims a lossless speedup to 5200 tps by integrating a diffusion adapter.