config.json

{
  "_attn_implementation_autoset": false,
  "add_cross_attention": false,
  "architectures": [
    "NemotronHForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 1,
  "chunk_size": 64,
  "conv_kernel": 4,
  "cross_attention_hidden_size": null,
  "decoder_start_token_id": null,
  "dtype": "bfloat16",
  "eos_token_id": 2,
  "expand": 2,
  "finetuning_task": null,
  "head_dim": 64,
  "hidden_dropout": 0.0,
  "hidden_size": 512,
  "initializer_range": 0.02,
  "intermediate_size": 256,
  "is_decoder": false,
  "layer_norm_epsilon": 1e-05,
  "layers_block_type": [
    "mamba",
    "moe",
    "mamba",
    "moe",
    "mamba",
    "attention"
  ],
  "mamba_head_dim": 32,
  "mamba_hidden_act": "silu",
  "mamba_num_heads": 16,
  "mamba_proj_bias": false,
  "mamba_ssm_cache_dtype": "float32",
  "max_position_embeddings": 262144,
  "mlp_bias": false,
  "mlp_hidden_act": "relu2",
  "model_type": "nemotron_h",
  "moe_intermediate_size": 256,
  "moe_latent_size": 256,
  "moe_shared_expert_intermediate_size": 512,
  "moe_shared_expert_overlap": true,
  "mtp_layers_block_type": [
    "attention",
    "moe"
  ],
  "n_group": 1,
  "n_groups": 8,
  "n_routed_experts": 4,
  "n_shared_experts": 1,
  "norm_eps": 1e-05,
  "norm_topk_prob": true,
  "num_attention_heads": 8,
  "num_experts_per_tok": 2,
  "num_key_value_heads": 2,
  "num_logits_to_keep": 1,
  "num_nextn_predict_layers": 1,
  "pad_token_id": 0,
  "partial_rotary_factor": 1.0,
  "prefix": null,
  "pruned_heads": {},
  "rescale_prenorm_residual": true,
  "residual_in_fp32": false,
  "rope_theta": 10000,
  "routed_scaling_factor": 2.5,
  "sep_token_id": null,
  "sliding_window": null,
  "ssm_state_size": 128,
  "task_specific_params": null,
  "tf_legacy_loss": false,
  "tie_encoder_decoder": false,
  "tie_word_embeddings": false,
  "time_step_floor": 0.0001,
  "time_step_limit": [
    0.0,
    {
      "__float__": "Infinity"
    }
  ],
  "time_step_max": 0.1,
  "time_step_min": 0.001,
  "tokenizer_class": null,
  "topk_group": 1,
  "torchscript": false,
  "transformers_version": "5.3.0",
  "use_bfloat16": false,
  "use_bias": false,
  "use_cache": true,
  "use_conv_bias": true,
  "use_mamba_kernels": false,
  "vocab_size": 131072
}