config.json

{
  "architectures": [
    "BantamForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "attention_head_groups": [
    {
      "head_dim": 128,
      "kv_heads": 1,
      "query_heads": 3
    },
    {
      "head_dim": 64,
      "kv_heads": 2,
      "query_heads": 6
    }
  ],
  "attn_logit_softcapping": 0.0,
  "bos_token_id": 1,
  "dtype": "bfloat16",
  "eos_token_id": 2,
  "final_logit_softcapping": 0.0,
  "head_dim": 85,
  "hidden_size": 768,
  "initializer_range": 0.02,
  "intermediate_size": 2304,
  "label_smoothing": 0.0,
  "layer_configs": [
    {
      "attention_head_groups": [
        {
          "head_dim": 64,
          "kv_heads": 3,
          "query_heads": 12
        }
      ],
      "intermediate_size": 2304,
      "num_attention_heads": 12,
      "num_key_value_heads": 3,
      "window": 128
    },
    {
      "attention_head_groups": [
        {
          "head_dim": 64,
          "kv_heads": 3,
          "query_heads": 12
        }
      ],
      "intermediate_size": 2304,
      "num_attention_heads": 12,
      "num_key_value_heads": 3,
      "window": 128
    },
    {
      "attention_head_groups": [
        {
          "head_dim": 64,
          "kv_heads": 3,
          "query_heads": 12
        }
      ],
      "intermediate_size": 2368,
      "num_attention_heads": 12,
      "num_key_value_heads": 3,
      "window": 128
    },
    {
      "attention_head_groups": [
        {
          "head_dim": 64,
          "kv_heads": 3,
          "query_heads": 12
        }
      ],
      "intermediate_size": 2400,
      "num_attention_heads": 12,
      "num_key_value_heads": 3
    },
    {
      "attention_head_groups": [
        {
          "head_dim": 80,
          "kv_heads": 2,
          "query_heads": 6
        },
        {
          "head_dim": 96,
          "kv_heads": 1,
          "query_heads": 3
        }
      ],
      "expert_type": "topk",
      "intermediate_size": 2432,
      "moe_aux_loss_weight": 0.01,
      "moe_capacity_factor": 1.05,
      "moe_drop_policy": "random",
      "moe_intermediate_size": 2432,
      "moe_router_jitter": 0.05,
      "moe_top_k": 2,
      "num_attention_heads": 9,
      "num_experts": 6,
      "num_key_value_heads": 3,
      "window": 256
    },
    {
      "attention_head_groups": [
        {
          "head_dim": 80,
          "kv_heads": 2,
          "query_heads": 6
        },
        {
          "head_dim": 96,
          "kv_heads": 1,
          "query_heads": 3
        }
      ],
      "intermediate_size": 2368,
      "num_attention_heads": 9,
      "num_key_value_heads": 3,
      "window": 256
    },
    {
      "attention_head_groups": [
        {
          "head_dim": 80,
          "kv_heads": 2,
          "query_heads": 6
        },
        {
          "head_dim": 96,
          "kv_heads": 1,
          "query_heads": 3
        }
      ],
      "intermediate_size": 2432,
      "num_attention_heads": 9,
      "num_key_value_heads": 3,
      "window": 256
    },
    {
      "attention_head_groups": [
        {
          "head_dim": 80,
          "kv_heads": 2,
          "query_heads": 6
        },
        {
          "head_dim": 96,
          "kv_heads": 1,
          "query_heads": 3
        }
      ],
      "intermediate_size": 2368,
      "num_attention_heads": 9,
      "num_key_value_heads": 3
    },
    {
      "intermediate_size": 2304,
      "window": 256
    },
    {
      "intermediate_size": 2368,
      "window": 256
    },
    {
      "intermediate_size": 2400,
      "window": 256
    },
    {
      "intermediate_size": 2432,
      "window": 256
    },
    {
      "intermediate_size": 2432
    },
    {
      "intermediate_size": 2400,
      "window": 512
    },
    {
      "intermediate_size": 2432,
      "window": 512
    },
    {
      "intermediate_size": 2432,
      "window": 512
    },
    {
      "expert_type": "topk",
      "intermediate_size": 2432,
      "moe_aux_loss_weight": 0.01,
      "moe_capacity_factor": 1.1,
      "moe_drop_policy": "random",
      "moe_intermediate_size": 2432,
      "moe_router_jitter": 0.05,
      "moe_top_k": 2,
      "num_experts": 8,
      "window": 512
    },
    {
      "expert_type": "topk",
      "intermediate_size": 2432,
      "moe_aux_loss_weight": 0.01,
      "moe_capacity_factor": 1.1,
      "moe_drop_policy": "random",
      "moe_intermediate_size": 2432,
      "moe_router_jitter": 0.05,
      "moe_top_k": 2,
      "num_experts": 8
    },
    {
      "intermediate_size": 2368,
      "window": 512
    },
    {
      "intermediate_size": 2400
    }
  ],
  "max_head_dim": 128,
  "max_position_embeddings": 2048,
  "mlp_dropout": 0.0,
  "model_type": "bantam",
  "num_attention_heads": 9,
  "num_attention_sinks": 4,
  "num_hidden_layers": 20,
  "num_key_value_heads": 3,
  "pad_token_id": 1,
  "qk_norm": true,
  "qk_norm_eps": 1e-06,
  "residual_dropout": 0.0,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 10000.0,
  "scaled_embeddings": false,
  "sink_boost": 0.25,
  "transformers_version": "4.57.1",
  "use_cache": false,
  "vocab_size": 49152,
  "z_loss_weight": 0.0
}