{ "architectures": [ "Qwen4ExpForCausalLM" ], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 248044, "dtype": "bfloat16", "eos_token_id": 248044, "hc_count": 4, "hc_lowrank": 192, "head_dim": 128, "heads_per_ngram": 8, "hidden_act": "silu", "hidden_size": 1536, "indexer_budget": 1024, "indexer_compress_ratio": 4, "indexer_head_dim": 64, "indexer_kv_heads": 1, "indexer_n_heads": 4, "initializer_range": 0.02, "layer_types": [ "linear_attention", "linear_attention", "linear_attention", "qwen_sparse_attention", "linear_attention", "linear_attention", "linear_attention", "qwen_sparse_attention", "linear_attention", "linear_attention", "linear_attention", "qwen_sparse_attention", "linear_attention", "linear_attention", "linear_attention", "qwen_sparse_attention", "linear_attention", "linear_attention", "linear_attention", "qwen_sparse_attention", "linear_attention", "linear_attention", "linear_attention", "qwen_sparse_attention", "linear_attention", "linear_attention", "linear_attention", "qwen_sparse_attention", "linear_attention", "linear_attention", "linear_attention", "qwen_sparse_attention" ], "linear_conv_kernel_dim": 4, "linear_key_head_dim": 128, "linear_num_key_heads": 8, "linear_num_value_heads": 24, "linear_value_head_dim": 128, "make_ngram_vocab_size_divisible_by": 128, "max_position_embeddings": 32768, "model_type": "qwen4_exp_text", "moe_intermediate_size": 768, "ngram_size": 3, "ngram_vocab_size_base": 144128, "norm_topk_prob": true, "num_attention_heads": 12, "num_experts": 64, "num_experts_per_tok": 2, "num_hidden_layers": 32, "num_key_value_heads": 2, "number_of_conv_states": 3, "output_gate_type": "sigmoid", "output_router_logits": false, "pad_token_id": 248044, "ple_conv_kernel_size": 4, "ple_embed_dim": 1536, "ple_layer_ids": [ 2 ], "rms_norm_eps": 1e-06, "rope_parameters": { "partial_rotary_factor": 0.25, "rope_theta": 10000000, "rope_type": "default" }, "router_aux_loss_coef": 0.001, "seed": 1234, "shared_expert_intermediate_size": 768, "split_ngram_parts": 128, "tie_word_embeddings": false, "transformers_version": "5.17.0", "use_cache": true, "vocab_size": 248320 }