AI · Research · 2026-11-10 22:44
Research reveals that directly deleting weights during LLM post-training significantly boosts performance, challenging traditional fine-tuning paradigms and offering a novel optimization strategy.
Auto-generated summary for reference only. RSS