Trying "DFlash," a Diffusion-Model Approach to Parallel Draft-Token Generation, on Gemma

Automated excerpt

The short version: DFlash did not outperform the Assistant model. Z-Lab's DFlash brings that diffusion-model property into the MTP draft model. Drafter model: gemma-4-12B-it-DFlash (https://huggingface.co/z-lab/gemma4-12B-it-DFlash) We use Z-Lab's DFlash model for Gemma-12B. The Assistant model shares its KV cache with the main Gemma model. If model tuning progresses further and a faster-processing diffusion model becomes possible, DFlash might eventually surpass the Assistant model.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.