@yepengliu
Ph.D. Student @ucsantabarbara @ucsbcs
AI
A post says TTME kept model weights fixed and used 82.8%–84.6% fewer FLOPs than test-time training on three tasks.
AI
