LeVJEPA 1.5 Large's creators say it closes the gap to V-JEPA on image and video benchmarks.
Its creators say LeVJEPA 1.5 Large was trained with 5.6 times less compute and over an order of magnitude less data while closing the gap to V-JEPA on image and video benchmarks. They call it a first step in scaling their method.
16.3K
5 Sources, first seen ago
@lukaskuhn77Introducing LeVJEPA 1.5 Large, the first step in scaling our method. It closes the gap to V-JEPA on both image and video benchmarks, trained with 5.6x less compute and over an order of magnitude less data.
@randall_balestrAnother example that we should spend more time thinking about improving pretraining losses rather than just searching for more data/gpus! SIGReg is not the final answer, but certainly a good first step in that direction! Try it out, checkpoint is available!
@soniajoseph_RT @lukaskuhn77: Introducing LeVJEPA 1.5 Large, the first step in scaling our method. It closes the gap to V-JEPA on both image and video…@lukaskuhn77Introducing LeVJEPA 1.5 Large, the first step in scaling our method. It closes the gap to V-JEPA on both image and video benchmarks, trained with 5.6x less compute and over an order of magnitude less data.@randall_balestrAnother example that we should spend more time thinking about improving pretraining losses rather than just searching for more data/gpus! SIGReg is not the final answer, but certainly a good first step in that direction! Try it out, checkpoint is available!@soniajoseph_RT @lukaskuhn77: Introducing LeVJEPA 1.5 Large, the first step in scaling our method. It closes the gap to V-JEPA on both image and video…@lukaskuhn77Introducing LeVJEPA 1.5 Large, the first step in scaling our method. It closes the gap to V-JEPA on both image and video benchmarks, trained with 5.6x less compute and over an order of magnitude less data.@randall_balestrAnother example that we should spend more time thinking about improving pretraining losses rather than just searching for more data/gpus! SIGReg is not the final answer, but certainly a good first step in that direction! Try it out, checkpoint is available!@soniajoseph_RT @lukaskuhn77: Introducing LeVJEPA 1.5 Large, the first step in scaling our method. It closes the gap to V-JEPA on both image and video…@lukaskuhn77Introducing LeVJEPA 1.5 Large, the first step in scaling our method. It closes the gap to V-JEPA on both image and video benchmarks, trained with 5.6x less compute and over an order of magnitude less data.@randall_balestrAnother example that we should spend more time thinking about improving pretraining losses rather than just searching for more data/gpus! SIGReg is not the final answer, but certainly a good first step in that direction! Try it out, checkpoint is available!@soniajoseph_RT @lukaskuhn77: Introducing LeVJEPA 1.5 Large, the first step in scaling our method. It closes the gap to V-JEPA on both image and video…—
Not ranked yet
—
Not ranked yet
