JEM trained in 20 H100-hours lands 1 point behind fine-tuned DINOv3
gabriberton · x · 2026-11-16
Researcher Simon Goodman shared results from training JEM (built on LeJEPA) for 100 epochs: using only 20 H100 hours on 250k images, it landed just 1 point behind a fine-tuned DINOv3, and he believes more training could beat it. The main issue is rare classes, likely a sampling problem. Resharing it, gabriberton praised the style: run simple experiments and share findings immediately—no novelty-seeking, no 50-page papers, no review queues.
More from Research
- Translated open-sources Lara Mini Base: 0.6B-active MoE beats models trained on 4-6x more tokens — NielsRogge · 2026-11-16
- Researchers formally specify ops interfaces for CockroachDB, Kafka and RabbitMQ to make AI operators verifiable — tianyin_xu · 2026-11-16
- Astera Residencies open: up to $2M in funding for off-network researchers — dr_alphalyrae · 2026-11-16
- 'It Knows Google Is Watching': Viral Chat Surfaces J-Lens Evaluation-Awareness Research — repligate · 2026-11-16
- NeurIPS Paper Explains Why Knowledge Editing in English Fails to Propagate Across Languages in Multilingual LLMs — Tanmoy_Chak · 2026-11-16
- MDPI review surveys reinforcement learning approaches in quantitative trading — PtrPomorski · 2026-11-16