{ "architectures": [ "MOSSv1d6ImageAsPrefixVLModel" ], "compressor_config": { "input_channels": 1024, "intermediate_size": 1536, "output_channels": 1024, "ratio": 2 }, "decoder_config": { "abs_pe": false, "abs_pe_max_length": 8192, "addition_eos_token_ls": null, "attention_dropout": 0.0, "attn_bias": true, "attn_o_bias": false, "bos_token_id": 1, "cross_attn_layers": [], "cross_attn_num_key_value_heads": 2, "eos_token_id": 2, "gradient_checkpointing": false, "head_dim": 128, "hidden_act": "silu", "hidden_size": 1024, "img_token_id": 67839, "initializer_range": 0.02, "intermediate_size": 1536, "kwargs": {}, "max_length": 4096, "max_position_embeddings": 8192, "model_path": null, "model_type": "", "no_bos_sampling": false, "num_attention_heads": 8, "num_hidden_layers": 12, "num_key_value_heads": 2, "output_cross_attentions": false, "partial_rope": false, "return_dict": true, "rms_norm_eps": 1e-06, "rope_theta": 1000000.0, "scale_embedding": false, "self_attn_layers": [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 ], "sliding_window": null, "use_qk_norm": true, "vocab_size": 67840 }, "dtype": "float32", "encoder_config": { "eps": 1e-05, "ffn_type": "swiglu", "gradient_checkpointing": false, "grid_size": [ 96, 96 ], "head_dim": 128, "height": 1536, "hidden_act": "silu", "hidden_size": 1024, "image_size": [ 1536, 1536 ], "intermediate_size": 1536, "merge_size": 2, "model_path": null, "model_type": "", "num_attention_heads": 8, "num_channels": 3, "num_hidden_layers": 12, "num_key_value_heads": 4, "partial_rope": false, "patch_size": 16, "qk_norm": false, "rope_theta": 10000.0, "stride": 16, "use_post_norm": false, "use_pre_norm": true, "use_qk_norm": true, "width": 1536 }, "linear_adapter": false, "model_type": "moss_v1d6", "transformers_version": "4.57.6", "auto_map": { "AutoConfig": "configuration_moss_v1d6.MossV1d6Config", "AutoModelForCausalLM": "modeling_moss_v1d6.MossV1d6ForCausalLM" } }