Cluster: finder
Stories below may also appear in other clusters. Data expires after 7 days.
GitHub - alainnothere / llm - circuit - finder : I replicated Ng RYS method and found that duplicating 3 specific layers in Qwen2 . 5 - 32B boosts reasoning by 17 % and duplicating layers 12 - 14 in Devstral - 24B improves logical deduction from 0 . ...