config.json

{
  "vocab_size": 248320,
  "hidden_size": 1024,
  "num_layers": 20,
  "full_attention_interval": 4,
  "global_attention_layers": [
    3,
    7,
    11,
    15,
    19
  ],
  "sliding_window": 256,
  "attn_q_heads": 16,
  "attn_kv_heads": 2,
  "attn_head_dim": 64,
  "rope_dim": 64,
  "rope_theta": 500000.0,
  "gdn_qk_heads": 8,
  "gdn_v_heads": 4,
  "gdn_head_dim": 64,
  "gdn_conv_kernel": 4,
  "gdn_chunk_size": 32,
  "num_experts": 8,
  "num_active_experts": 2,
  "num_shared_experts": 1,
  "expert_intermediate": 2176,
  "router_aux_loss_coef": 0.01,
  "router_z_loss_coef": 0.001,
  "expert_score_band": 0.15,
  "use_mod": true,
  "mod_top_k": 3,
  "mod_threshold": 0.15,
  "mod_aux_loss_coef": 0.01,
  "train_seq_len": 2048,
  "rms_eps": 1e-06,
  "dtype": "bfloat16",
  "grad_checkpoint_segment": 1
}