conversion_info.json

{
  "method": "drop_upcycling",
  "drop_ratio": 0.66,
  "num_experts_per_layer": 288,
  "base_experts_from_ffn": 12,
  "replication_factor": 24,
  "total_experts": 8064,
  "source_ffn_dim": 8960,
  "target_expert_dim": 768,
  "layers": 28,
  "total_params": 28938036224,
  "total_params_calculated": 28860965888,
  "active_params_per_forward": 917464064,
  "activation_ratio": 0.03170443415365876,
  "expert_params": 3538944,
  "total_expert_params": 28538044416
}