{"transformers_version": "5.16.1", "architectures": ["GrugMoeForCausalLM"], "output_hidden_states": false, "return_dict": true, "dtype": "bfloat16", "chunk_size_feed_forward": 0, "is_encoder_decoder": false, "id2label": {"0": "LABEL_0", "1": "LABEL_1"}, "label2id": {"LABEL_0": 0, "LABEL_1": 1}, "problem_type": null, "_name_or_path": "", "vocab_size": 128256, "hidden_size": 2560, "num_hidden_layers": 26, "num_attention_heads": 20, "num_key_value_heads": 5, "head_dim": 128, "max_position_embeddings": 262144, "sliding_window": 2048, "rms_norm_eps": 1e-05, "initializer_range": 0.009882117688026186, "rope_theta": 10000.0, "tie_word_embeddings": false, "num_experts": 256, "num_experts_per_tok": 4, "moe_intermediate_size": 1280, "shared_expert_intermediate_size": 2560, "qk_mult": 1.75, "grugmoe_attention_mode": "production", "grugmoe_artifact_schema_version": 1, "model_type": "grug_moe", "output_attentions": false, "pad_token_id": 128001, "eos_token_id": 128001, "bos_token_id": 128000, "decoder_start_token_id": 128000, "begin_suppress_tokens": [128000, 128001], "disable_long_rope": true, "disable_pko": true, "global_every": 4, "global_kv_heads": null, "latent_dim": null, "local_kv_heads": null, "num_shared_experts": 1, "qk_mult_long_scale": 1.0, "rope_fused": false, "router_z_loss_coef": 0.0, "sconv": false, "sconv_kernel": 4, "sconv_sites": ["k", "attn", "mlp"]}