ByteDance released an open-source model that attempts broad multimodal tasks with 3B parameters
bytedance released an open source model that attempts to do just about anything with only 3b parameters
ByteDance released Lance, an open-source unified multimodal model with 3B active parameters that supports image and video understanding, generation, and editing, and the post says it was trained from scratch with a staged multi-task recipe under a 128-A100-GPU budget.
Why it matters: ByteDance open-sourced Lance, which puts image and video understanding, generation and editing in one model with 3B active parameters, plus the staged multi-task training recipe and 128-A100 setup.