Open-Source Fast-Float Pushes Talking Avatar Inference to 90 FPS on a T4

Developer mohamed-em2m open-sourced Fast-Float, a self-hosted talking-avatar inference server that optimizes FLOAT model performance from 30 to 90 FPS on a single free Colab T4 GPU, demonstrated with a Messi avatar.

2026-10-29 ~ 2026-10-29 · 2 related posts

• Self-hosted realtime avatar server: 30 FPS to 90 FPS on a free Colab T4 — SavingsWeather1659 · 2026-10-29
• Fast-Float pushes FLOAT avatar inference to 90 FPS on a free T4, 105 FPS with TensorRT — SavingsWeather1659 · 2026-10-29