DiffusionGemma: 4배 빠른 텍스트 생성

DiffusionGemma introduces parallel text generation through diffusion-based decoding, delivering up to 4x faster inference on consumer GPUs for local deployments compared to traditional autoregressive models. This shift from sequential token generation to simultaneous paragraph-level generation makes real-time, offline AI interactions feasible on resource-constrained devices. The trade-off is lower output quality, positioning it as optimal for speed-critical, interactive workflows.