Announcement
Turning DiffusionGemma into a Jev-like model with vLLM
Google Gemma says seeding a canvas with a response template lets vLLM extract confidence scores and probability distributions for yes/no, multiple-choice and scored questions in one denoising step.
TLDR
Google Gemma describes a way to turn DiffusionGemma into a Jev-like model using vLLM. It says a response template seeded on a canvas enables confidence scores and probability distributions for structured questions in one denoising step.
Combined views
63.6K
2 Sources, first seen 6h ago
1.1K likes