{ "architectures": [ "Lfm2DSparkDraftModel" ], "model_type": "qwen3", "dtype": "bfloat16", "hidden_size": 2048, "num_hidden_layers": 5, "num_attention_heads": 32, "num_key_value_heads": 8, "head_dim": 64, "intermediate_size": 6144, "hidden_act": "silu", "rms_norm_eps": 1e-05, "vocab_size": 128000, "rope_theta": 10000000.0, "max_position_embeddings": 128000, "layer_types": [ "full_attention", "full_attention", "full_attention", "full_attention", "full_attention" ], "block_size": 9, "dflash_config": { "mask_token_id": 125017, "target_layer_ids": [ 2, 9, 17, 21, 27 ], "num_target_layers": 30 }, "markov_rank": 256, "rope_is_neox_style": false, "enable_confidence_head": true, "markov_head_type": "vanilla" }