LLaDA2.0-flash-preview / config.json
lccurious's picture
First model version
b359dde
{
"architectures": [
"LLaDA2MoeModelLM"
],
"attention_dropout": 0.0,
"auto_map": {
"AutoConfig": "configuration_llada2_moe.LLaDA2MoeConfig",
"AutoModel": "modeling_llada2_moe.LLaDA2MoeModel",
"AutoModelForCausalLM": "modeling_llada2_moe.LLaDA2MoeModelLM"
},
"num_hidden_layers": 32,
"hidden_size": 4096,
"intermediate_size": 9216,
"first_k_dense_replace": 1,
"hidden_act": "silu",
"max_position_embeddings": 16384,
"model_type": "llada2_moe",
"moe_intermediate_size": 1024,
"norm_topk_prob": true,
"num_experts_per_tok": 8,
"norm_head": false,
"num_attention_heads": 32,
"num_experts": 256,
"num_key_value_heads": 4,
"rope_theta": 600000,
"rope_scaling": null,
"tie_word_embeddings": false,
"torch_dtype": "bfloat16",
"transformers_version": "4.52.3",
"use_bias": false,
"use_rmsnorm": true,
"rms_norm_eps": 1e-06,
"head_dim": 128,
"num_shared_experts": 1,
"use_cache": true,
"use_qkv_bias": false,
"embedding_dropout": 0.0,
"norm_softmax": false,
"output_dropout": 0.0,
"vocab_size": 157184,
"rotary_dim": 64,
"using_split_qkv_in_self_attention": false,
"router_dtype": "fp32",
"moe_router_enable_expert_bias": true,
"routed_scaling_factor": 2.5,
"n_group": 8,
"topk_group": 4,
"score_function": "sigmoid",
"initializer_range": 0.02,
"max_window_layers": 28,
"output_router_logits": false,
"pad_token_id": 156892,
"partial_rotary_factor": 0.5,
"use_sliding_window": false
}