iter74: add num_local_experts/num_shared_experts aliases (validator MoE accounting) 4a86abc verified unconst commited on 20 days ago
iter67 souped_slerp_a30 SLERP alpha=0.30 (hope_king<->new_king) merged via merge_slerp.py; probe top1=0.5661 82644f0 verified unconst commited on 20 days ago