{
  "_comment": "Ultra-deep reasoning config: 24L/12H/768d, mlp_mult=4. Maximum reasoning depth for RAG + multi-step inference. ~11-12GB VRAM.",
  "architectures": [
    "VeronicaForCausalLM"
  ],
  "bos_token_id": 50257,
  "dropout": 0.0,
  "dtype": "bfloat16",
  "eos_token_id": 50256,
  "gradient_checkpointing": true,
  "hidden_size": 768,
  "layer_norm_epsilon": 1e-05,
  "max_position_embeddings": 2048,
  "mlp_mult": 4,
  "model_type": "veronica",
  "n_embd": 768,
  "n_head": 12,
  "n_layer": 24,
  "num_attention_heads": 12,
  "num_funcs": 3,
  "num_hidden_layers": 24,
  "pad_token_id": 50258,
  "rope_theta": 10000.0,
  "router_aux_weight": 0.015999763124389305,
  "router_dim": 128,
  "router_tau": 1.400028804954452,
  "transformers_version": "4.57.0.dev0",
  "use_cache": false,
  "use_channel_attention": false,
  "use_flash_attn": true,
  "vocab_size": 50271
}