LynnReal-Omni-Flash
A lightweight, accelerated variant of LynnReal-Omni with 27 billion parameters. It uses a three-step generation process and a lightweight VAE decoder, optimized for low-latency real-time rendering scenarios.
Timeline (last 90 days)
The team trained a 27B-parameter LynnReal-Omni-Flash version that uses three-step generation to reduce inference overhead through model acceleration, decoding acceleration, and a lightweight VAE decoder.
A lightweight VAE decoder may produce slight fidelity loss on extreme texture details or high-frequency edges.
All Facts (2)
A lightweight VAE decoder may produce slight fidelity loss on extreme texture details or high-frequency edges.
50%UnverifiedThe team trained a 27B-parameter LynnReal-Omni-Flash version that uses three-step generation to reduce inference overhead through model acceleration, decoding acceleration, and a lightweight VAE decoder.
50%