iter72: add num_local_experts/num_shared_experts aliases (validator MoE accounting) 09bfd61 verified unconst commited on 20 days ago
iter65 souped_slerp_a50 SLERP alpha=0.50 (hope_king<->new_king) merged via merge_slerp.py; probe top1=0.5521 fb8a6df verified unconst commited on 20 days ago