config.json

{
  "architectures": [
    "GraniteMoeHybridForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "attention_multiplier": 1.0,
  "bos_token_id": 1,
  "embedding_multiplier": 1.0,
  "eos_token_id": 2,
  "hidden_act": "silu",
  "hidden_size": 32,
  "initializer_range": 0.02,
  "intermediate_size": 16,
  "layer_types": [
    "mamba",
    "mamba",
    "attention",
    "mamba"
  ],
  "logits_scaling": 1.0,
  "mamba_chunk_size": 16,
  "mamba_conv_bias": true,
  "mamba_d_conv": 3,
  "mamba_d_head": 8,
  "mamba_d_state": 16,
  "mamba_expand": 2,
  "mamba_n_groups": 1,
  "mamba_n_heads": 8,
  "mamba_proj_bias": false,
  "max_position_embeddings": 2048,
  "model_type": "granitemoehybrid",
  "num_attention_heads": 32,
  "num_experts_per_tok": 2,
  "num_heads": 8,
  "num_hidden_layers": 4,
  "num_key_value_heads": 32,
  "num_local_experts": 4,
  "output_router_logits": false,
  "position_embedding_type": null,
  "residual_multiplier": 1.0,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 10000.0,
  "router_aux_loss_coef": 0.001,
  "shared_intermediate_size": 1024,
  "tie_word_embeddings": false,
  "torch_dtype": "float32",
  "transformers_version": "4.55.4",
  "use_cache": true,
  "vocab_size": 100352
}