iter73: add num_local_experts/num_shared_experts aliases (validator MoE accounting) e905e14 verified unconst commited on 21 days ago
iter68 souped_slerp_a20 SLERP alpha=0.20 (hope_king<->new_king) merged via merge_slerp.py; probe top1=0.5677 64d9539 verified unconst commited on 22 days ago