config.json

{
  "add_tail_layers": false,
  "architectures": [
    "Ernie4_5_VLMoeForConditionalGeneration"
  ],
  "attention_probs_dropout_prob": 0.0,
  "auto_map": {
    "AutoConfig": "baidu/ERNIE-4.5-VL-424B-A47B-PT--configuration_ernie4_5_vl.Ernie4_5_VLMoEConfig",
    "AutoImageProcessor": "baidu/ERNIE-4.5-VL-424B-A47B-PT--processing_ernie4_5_vl.Ernie4_5_VLImageProcessor",
    "AutoModel": "baidu/ERNIE-4.5-VL-424B-A47B-PT--modeling_ernie4_5_vl.Ernie4_5_VLMoeForConditionalGeneration",
    "AutoModelForCausalLM": "baidu/ERNIE-4.5-VL-424B-A47B-PT--modeling_ernie4_5_vl.Ernie4_5_VLMoeForConditionalGeneration",
    "AutoProcessor": "baidu/ERNIE-4.5-VL-424B-A47B-PT--processing_ernie4_5_vl.Ernie4_5_VLProcessor"
  },
  "bos_token_id": 1,
  "cachekv_quant": false,
  "compression_ratio": 1.0,
  "disable_ffn_model_parallel": false,
  "dpo_config": null,
  "dtype": "bfloat16",
  "enable_delay_scale_loss": true,
  "eos_token_id": 2,
  "freq_allocation": 20,
  "fuse_attn_ffn": true,
  "fuse_gate_detach_matmul": false,
  "fuse_linear": false,
  "fuse_ln": false,
  "fuse_rms_norm": false,
  "fuse_rope": false,
  "fuse_softmax_mask": false,
  "fuse_swiglu": false,
  "global_aux_loss": false,
  "hidden_act": "silu",
  "hidden_dropout_prob": 0.0,
  "hidden_size": 8,
  "ignored_index": -100,
  "im_patch_id": 100295,
  "initializer_range": 0.20411393876950196,
  "intermediate_size": 32,
  "loss_subbatch_seqlen": 8192,
  "max_position_embeddings": 131072,
  "max_sequence_length": null,
  "max_text_id": null,
  "micro_batch_size": -1,
  "mm_vocab_size": 0,
  "modality_detach": false,
  "model_type": "ernie4_5_moe_vl",
  "moe_all_to_all_dropout": 0.0,
  "moe_aux_loss_lambda": 0.01,
  "moe_capacity": [
    64,
    64,
    64
  ],
  "moe_dense_experts_token_type_id": 3,
  "moe_dropout_prob": 0.0,
  "moe_fuse_experts": false,
  "moe_gate": "topk",
  "moe_gate_act": "softmax",
  "moe_group": "world",
  "moe_group_experts": false,
  "moe_group_orthogonal_loss": true,
  "moe_intermediate_size": [
    32,
    32
  ],
  "moe_k": 8,
  "moe_layer_end_index": 1,
  "moe_layer_feed_fake_token": false,
  "moe_layer_interval": 1,
  "moe_layer_start_index": 1,
  "moe_multimodal_dispatch_use_allgather": "v2-alltoall-unpad-text",
  "moe_norm_gate_logits": true,
  "moe_num_attn_experts": false,
  "moe_num_experts": [
    32,
    32
  ],
  "moe_num_shared_experts": 0,
  "moe_orthogonal_loss_lambda": 0.01,
  "moe_reverse_token_drop": false,
  "moe_use_aux_free": true,
  "moe_use_hard_gate": true,
  "moe_use_size_all2all": false,
  "moe_use_token_type_bias": false,
  "moe_z_loss_lambda": 0.0001,
  "num_acc_steps": 1,
  "num_attention_heads": 4,
  "num_hidden_layers": 2,
  "num_key_value_heads": 4,
  "output_attentions": false,
  "pad_token_id": 0,
  "pixel_hidden_size": 16,
  "pp_seg_method": "layer:ErnieDecoderLayer|EmptyLayer",
  "recompute": false,
  "recompute_granularity": "core_attn",
  "recompute_use_reentrant": false,
  "refined_recompute": {},
  "resampler_fuse_rms_norm": false,
  "rms_norm_eps": 1e-05,
  "rope_3d": true,
  "rope_scaling": {
    "mrope_section": [
      22,
      22,
      20
    ],
    "type": "default"
  },
  "rope_theta": 500000,
  "sinkhorn_2gate": true,
  "sinkhorn_temp": 0.03,
  "skip_recompute_ops": {},
  "spatial_conv_size": 2,
  "temporal_conv_size": 2,
  "tensor_parallel_degree": 1,
  "tie_word_embeddings": false,
  "token_balance_loss": false,
  "token_balance_seqlen": false,
  "transformers_version": null,
  "use_bias": false,
  "use_cache": true,
  "use_ep_comm_overlap": false,
  "use_fast_ln": false,
  "use_flash_attention": true,
  "use_fused_head_and_loss_fn": false,
  "use_recompute_lm_head": false,
  "use_recompute_loss_fn": false,
  "use_recompute_moe": false,
  "use_recompute_resampler": false,
  "use_rmsnorm": true,
  "use_sparse_flash_attn": true,
  "use_sparse_head_and_loss_fn": false,
  "use_temporal_conv": true,
  "use_var_len_flash_attn": false,
  "using_precision_check": false,
  "video_end_token_id": 101307,
  "video_start_token_id": 101306,
  "vision_config": {
    "attn_implementation": "eager",
    "attn_sep": true,
    "depth": 2,
    "embed_dim": 16,
    "hidden_act": "quick_gelu",
    "hidden_size": 16,
    "in_channels": 3,
    "in_chans": 3,
    "mlp_ratio": 4,
    "model_type": "DFNRope_vision_transformer",
    "num_heads": 1,
    "patch_size": 14,
    "pp_data_balance": false,
    "recompute": false,
    "spatial_merge_size": 2,
    "spatial_patch_size": 14,
    "vit_first_fwd_bsz": 128,
    "vit_num_recompute_layers": 10000
  },
  "vocab_size": 103424,
  "weight_share_add_bias": true
}