{
  "architecture": {
    "attention_variant": "MHA",
    "family": "image_text_to_text",
    "is_moe": false,
    "mixer": "attention",
    "positional": "learned",
    "tie_word_embeddings": true,
    "towers": [
      {
        "attention_variant": "MHA",
        "config_class": "BlipTextConfig",
        "hidden_size": 768,
        "name": "text",
        "num_attention_heads": 8,
        "num_hidden_layers": 12,
        "positional": "learned"
      },
      {
        "attention_variant": "MHA",
        "config_class": "BlipVisionConfig",
        "hidden_size": 768,
        "name": "vision",
        "num_attention_heads": 12,
        "num_hidden_layers": 12
      }
    ],
    "view": "multimodal"
  },
  "capabilities": {
    "attention_backends": [
      "eager"
    ],
    "attention_patterns": [
      "causal"
    ],
    "attention_schedule": null,
    "task_heads": [
      "image_text_to_text",
      "multimodal_lm",
      "visual_question_answering",
      "zero_shot_image_classification"
    ],
    "tensor_parallel": false
  },
  "components": [
    {
      "children": [
        "text_model",
        "vision_model",
        "visual_projection",
        "text_projection"
      ],
      "class_name": "BlipModel",
      "id": "model",
      "kind": "model",
      "path_pattern": "model"
    },
    {
      "children": [
        "text_model.embeddings",
        "text_model.encoder",
        "text_model.pooler"
      ],
      "class_name": "BlipTextModel",
      "id": "text_model",
      "kind": "model",
      "path_pattern": "model.text_model"
    },
    {
      "attributes": {
        "embedding_dim": "config.hidden_size",
        "num_embeddings": "config.vocab_size"
      },
      "children": [
        "text_model.embeddings.word_embeddings",
        "text_model.embeddings.position_embeddings",
        "text_model.embeddings.LayerNorm"
      ],
      "class_name": "BlipTextEmbeddings",
      "id": "text_model.embeddings",
      "kind": "embedding",
      "path_pattern": "model.text_model.embeddings"
    },
    {
      "attributes": {
        "embedding_dim": "config.hidden_size",
        "num_embeddings": "config.vocab_size",
        "tied_lm_head": true
      },
      "class_name": "Embedding",
      "id": "text_model.embeddings.word_embeddings",
      "kind": "embedding",
      "path_pattern": "model.text_model.embeddings.word_embeddings"
    },
    {
      "attributes": {
        "max_position_embeddings": "config.max_position_embeddings",
        "scheme": "learned"
      },
      "class_name": "Embedding",
      "id": "text_model.embeddings.position_embeddings",
      "kind": "position",
      "path_pattern": "model.text_model.embeddings.position_embeddings"
    },
    {
      "attributes": {
        "norm_type": "layer"
      },
      "class_name": "LayerNorm",
      "id": "text_model.embeddings.LayerNorm",
      "kind": "normalization",
      "path_pattern": "model.text_model.embeddings.LayerNorm"
    },
    {
      "children": [
        "encoder_layers"
      ],
      "class_name": "BlipTextEncoder",
      "id": "text_model.encoder",
      "kind": "encoder",
      "path_pattern": "model.text_model.encoder"
    },
    {
      "class_name": "BlipTextPooler",
      "id": "text_model.pooler",
      "kind": "pooler",
      "path_pattern": "model.text_model.pooler"
    },
    {
      "children": [
        "vision_model.embeddings",
        "vision_model.encoder",
        "vision_model.post_layernorm"
      ],
      "class_name": "BlipVisionModel",
      "id": "vision_model",
      "kind": "model",
      "path_pattern": "model.vision_model"
    },
    {
      "attributes": {
        "embedding_dim": "config.hidden_size"
      },
      "children": [
        "vision_model.embeddings.patch_embedding"
      ],
      "class_name": "BlipVisionEmbeddings",
      "id": "vision_model.embeddings",
      "kind": "embedding",
      "path_pattern": "model.vision_model.embeddings"
    },
    {
      "class_name": "Conv2d",
      "id": "vision_model.embeddings.patch_embedding",
      "kind": "convolution",
      "path_pattern": "model.vision_model.embeddings.patch_embedding"
    },
    {
      "children": [
        "encoder_layer_2s"
      ],
      "class_name": "BlipEncoder",
      "id": "vision_model.encoder",
      "kind": "encoder",
      "path_pattern": "model.vision_model.encoder"
    },
    {
      "attributes": {
        "norm_type": "layer"
      },
      "class_name": "LayerNorm",
      "id": "vision_model.post_layernorm",
      "kind": "normalization",
      "path_pattern": "model.vision_model.post_layernorm"
    },
    {
      "attributes": {
        "in_features": "config.hidden_size",
        "out_features": 512
      },
      "class_name": "Linear",
      "id": "visual_projection",
      "kind": "projection",
      "path_pattern": "model.visual_projection"
    },
    {
      "attributes": {
        "in_features": 768,
        "out_features": 512
      },
      "class_name": "Linear",
      "id": "text_projection",
      "kind": "projection",
      "path_pattern": "model.text_projection"
    }
  ],
  "config": {
    "class_name": "BlipConfig",
    "model_type": "blip",
    "module": "transformers.models.blip.configuration_blip",
    "referenced_fields": {},
    "salient_fields": {
      "is_encoder_decoder": false,
      "tie_word_embeddings": true
    }
  },
  "dataflow": {
    "input": {
      "name": "input_ids",
      "shape": [
        "B",
        "S"
      ]
    },
    "output": {
      "shape": [
        "B",
        1
      ]
    },
    "shapes": {
      "encoder_layer.attention": {
        "in": [
          "B",
          "S",
          768
        ],
        "out": [
          "B",
          "S",
          768
        ]
      },
      "encoder_layer.intermediate": {
        "in": [
          "B",
          "S",
          768
        ],
        "out": [
          "B",
          "S",
          3072
        ]
      },
      "encoder_layer.output": {
        "in": [
          "B",
          "S",
          3072
        ],
        "out": [
          "B",
          "S",
          768
        ]
      },
      "encoder_layer_2.layer_norm1": {
        "in": [
          "B",
          577,
          768
        ],
        "out": [
          "B",
          577,
          768
        ]
      },
      "encoder_layer_2.layer_norm2": {
        "in": [
          "B",
          577,
          768
        ],
        "out": [
          "B",
          577,
          768
        ]
      },
      "encoder_layer_2.mlp": {
        "in": [
          "B",
          577,
          768
        ],
        "out": [
          "B",
          577,
          768
        ]
      },
      "encoder_layer_2.self_attn": {
        "in": null,
        "out": [
          "B",
          577,
          768
        ]
      },
      "text_model": {
        "in": null,
        "out": [
          "B",
          "S",
          768
        ]
      },
      "text_projection": {
        "in": [
          "B",
          768
        ],
        "out": [
          "B",
          512
        ]
      },
      "vision_model": {
        "in": null,
        "out": [
          "B",
          577,
          768
        ]
      },
      "visual_projection": {
        "in": [
          "B",
          768
        ],
        "out": [
          "B",
          512
        ]
      }
    },
    "source": "observed_forward_meta"
  },
  "edges": [
    {
      "kind": "data",
      "source": "text_model",
      "target": "vision_model"
    },
    {
      "kind": "data",
      "source": "text_model.embeddings",
      "target": "text_model.encoder"
    },
    {
      "kind": "data",
      "source": "text_model.encoder",
      "target": "text_model.pooler"
    },
    {
      "kind": "data",
      "source": "text_model.embeddings.word_embeddings",
      "target": "text_model.embeddings.LayerNorm"
    },
    {
      "kind": "data",
      "source": "encoder_layer.attention",
      "target": "encoder_layer.crossattention"
    },
    {
      "kind": "data",
      "source": "encoder_layer.crossattention",
      "target": "encoder_layer.intermediate"
    },
    {
      "kind": "data",
      "source": "encoder_layer.crossattention.self",
      "target": "encoder_layer.crossattention.output"
    },
    {
      "kind": "data",
      "source": "vision_model.embeddings",
      "target": "vision_model.encoder"
    },
    {
      "kind": "data",
      "source": "vision_model.encoder",
      "target": "vision_model.post_layernorm"
    },
    {
      "kind": "mask",
      "source": "input:attention_mask",
      "target": "encoder_layer.attention"
    },
    {
      "kind": "mask",
      "source": "input:attention_mask",
      "target": "encoder_layer.attention.self"
    },
    {
      "kind": "mask",
      "source": "input:attention_mask",
      "target": "encoder_layer.crossattention"
    },
    {
      "kind": "mask",
      "source": "input:attention_mask",
      "target": "encoder_layer.crossattention.self"
    },
    {
      "kind": "mask",
      "source": "input:attention_mask",
      "target": "encoder_layer.crossattention.output"
    },
    {
      "kind": "mask",
      "source": "input:attention_mask",
      "target": "encoder_layer.crossattention.output.LayerNorm"
    },
    {
      "kind": "mask",
      "source": "input:attention_mask",
      "target": "encoder_layer_2.self_attn"
    },
    {
      "kind": "residual",
      "source": "encoder_layer",
      "target": "encoder_layer.attention"
    },
    {
      "kind": "residual",
      "source": "encoder_layer",
      "target": "encoder_layer.crossattention"
    },
    {
      "kind": "residual",
      "source": "encoder_layer",
      "target": "encoder_layer.intermediate"
    },
    {
      "kind": "residual",
      "source": "encoder_layer_2",
      "target": "encoder_layer_2.self_attn"
    },
    {
      "kind": "residual",
      "source": "encoder_layer_2",
      "target": "encoder_layer_2.mlp"
    },
    {
      "kind": "position",
      "source": "text_model.embeddings.position_embeddings",
      "target": "text_model.embeddings"
    },
    {
      "kind": "cross_attention",
      "source": "text_model.encoder",
      "target": "encoder_layer.crossattention"
    },
    {
      "kind": "cross_attention",
      "source": "text_model.encoder",
      "target": "encoder_layer.crossattention.self"
    },
    {
      "kind": "cross_attention",
      "source": "text_model.encoder",
      "target": "encoder_layer.crossattention.output"
    },
    {
      "kind": "cross_attention",
      "source": "text_model.encoder",
      "target": "encoder_layer.crossattention.output.LayerNorm"
    },
    {
      "kind": "data",
      "provenance": "intra_module",
      "source": "encoder_layer.attention.self",
      "target": "encoder_layer.attention.self.query"
    },
    {
      "kind": "data",
      "provenance": "intra_module",
      "source": "encoder_layer.attention.self",
      "target": "encoder_layer.attention.self.key"
    },
    {
      "kind": "data",
      "provenance": "intra_module",
      "source": "encoder_layer.attention.self",
      "target": "encoder_layer.attention.self.value"
    },
    {
      "kind": "data",
      "provenance": "intra_module",
      "source": "encoder_layer.intermediate",
      "target": "encoder_layer.intermediate.dense"
    },
    {
      "kind": "data",
      "provenance": "intra_module",
      "source": "encoder_layer.output",
      "target": "encoder_layer.output.dense"
    },
    {
      "kind": "data",
      "provenance": "intra_module",
      "source": "encoder_layer_2.self_attn",
      "target": "encoder_layer_2.self_attn.qkv"
    },
    {
      "kind": "data",
      "provenance": "intra_module",
      "source": "encoder_layer_2.self_attn",
      "target": "encoder_layer_2.self_attn.projection"
    },
    {
      "kind": "data",
      "provenance": "intra_module",
      "source": "encoder_layer_2.mlp",
      "target": "encoder_layer_2.mlp.fc1"
    },
    {
      "kind": "data",
      "provenance": "intra_module",
      "source": "encoder_layer_2.mlp.fc1",
      "target": "encoder_layer_2.mlp.fc2"
    },
    {
      "kind": "data",
      "provenance": "observed_forward",
      "source": "encoder_layer.attention",
      "target": "encoder_layer.intermediate"
    },
    {
      "kind": "data",
      "provenance": "observed_forward",
      "source": "encoder_layer.intermediate",
      "target": "encoder_layer.output"
    },
    {
      "kind": "data",
      "provenance": "observed_forward",
      "source": "encoder_layer_2.layer_norm1",
      "target": "encoder_layer_2.self_attn"
    },
    {
      "kind": "data",
      "provenance": "observed_forward",
      "source": "encoder_layer_2.self_attn",
      "target": "encoder_layer_2.layer_norm2"
    },
    {
      "kind": "data",
      "provenance": "observed_forward",
      "source": "encoder_layer_2.layer_norm2",
      "target": "encoder_layer_2.mlp"
    },
    {
      "kind": "data",
      "provenance": "observed_forward",
      "source": "vision_model",
      "target": "text_model"
    },
    {
      "kind": "data",
      "provenance": "observed_forward",
      "source": "text_model",
      "target": "visual_projection"
    },
    {
      "kind": "data",
      "provenance": "observed_forward",
      "source": "visual_projection",
      "target": "text_projection"
    }
  ],
  "extends": null,
  "model_type": "blip",
  "provenance": {
    "config_class": "BlipConfig",
    "config_module": "transformers.models.blip.configuration_blip",
    "model_class": "BlipModel",
    "model_module": "transformers.models.blip.modeling_blip"
  },
  "repeats": [
    {
      "body": "encoder_layer",
      "container_path_pattern": "model.text_model.encoder.layer",
      "count": 12,
      "count_expr": "12",
      "count_source": "module_tree",
      "id": "encoder_layers",
      "index_symbol": "i",
      "item_path_pattern": "model.text_model.encoder.layer.{i}",
      "kind": "symbolic_repeat",
      "provenance": {
        "class_name": "BlipTextLayer",
        "container_path_pattern": "model.text_model.encoder.layer",
        "item_path_pattern": "model.text_model.encoder.layer.{i}",
        "source": "module_tree_repeat_collapse"
      },
      "repeated_class_name": "BlipTextLayer"
    },
    {
      "body": "encoder_layer_2",
      "container_path_pattern": "model.vision_model.encoder.layers",
      "count": 12,
      "count_expr": "12",
      "count_source": "module_tree",
      "id": "encoder_layer_2s",
      "index_symbol": "i",
      "item_path_pattern": "model.vision_model.encoder.layers.{i}",
      "kind": "symbolic_repeat",
      "provenance": {
        "class_name": "BlipEncoderLayer",
        "container_path_pattern": "model.vision_model.encoder.layers",
        "item_path_pattern": "model.vision_model.encoder.layers.{i}",
        "source": "module_tree_repeat_collapse"
      },
      "repeated_class_name": "BlipEncoderLayer"
    }
  ],
  "schema_version": "architecture-template-v0",
  "templates": [
    {
      "children": [
        "encoder_layer.attention",
        "encoder_layer.crossattention",
        "encoder_layer.intermediate",
        "encoder_layer.output"
      ],
      "class_name": "BlipTextLayer",
      "id": "encoder_layer",
      "kind": "transformer_block",
      "path_pattern": "model.text_model.encoder.layer.{i}"
    },
    {
      "attributes": {
        "head_dim": 96,
        "n_heads": 8,
        "n_kv_heads": 8,
        "rope": false,
        "variant": "MHA"
      },
      "children": [
        "encoder_layer.attention.self",
        "encoder_layer.attention.output"
      ],
      "class_name": "BlipTextAttention",
      "id": "encoder_layer.attention",
      "kind": "attention",
      "path_pattern": "model.text_model.encoder.layer.{i}.attention"
    },
    {
      "attributes": {
        "head_dim": 96,
        "n_heads": 8,
        "n_kv_heads": 8,
        "rope": false,
        "variant": "MHA"
      },
      "children": [
        "encoder_layer.attention.self.query",
        "encoder_layer.attention.self.key",
        "encoder_layer.attention.self.value"
      ],
      "class_name": "BlipTextSelfAttention",
      "id": "encoder_layer.attention.self",
      "kind": "attention",
      "path_pattern": "model.text_model.encoder.layer.{i}.attention.self"
    },
    {
      "attributes": {
        "in_features": "config.hidden_size",
        "out_features": "config.hidden_size"
      },
      "class_name": "Linear",
      "id": "encoder_layer.attention.self.query",
      "kind": "projection",
      "path_pattern": "model.text_model.encoder.layer.{i}.attention.self.query"
    },
    {
      "attributes": {
        "in_features": "config.hidden_size",
        "out_features": "config.hidden_size"
      },
      "class_name": "Linear",
      "id": "encoder_layer.attention.self.key",
      "kind": "projection",
      "path_pattern": "model.text_model.encoder.layer.{i}.attention.self.key"
    },
    {
      "attributes": {
        "in_features": "config.hidden_size",
        "out_features": "config.hidden_size"
      },
      "class_name": "Linear",
      "id": "encoder_layer.attention.self.value",
      "kind": "projection",
      "path_pattern": "model.text_model.encoder.layer.{i}.attention.self.value"
    },
    {
      "children": [
        "encoder_layer.attention.output.LayerNorm"
      ],
      "class_name": "BlipTextSelfOutput",
      "id": "encoder_layer.attention.output",
      "kind": "module",
      "path_pattern": "model.text_model.encoder.layer.{i}.attention.output"
    },
    {
      "attributes": {
        "norm_type": "layer"
      },
      "class_name": "LayerNorm",
      "id": "encoder_layer.attention.output.LayerNorm",
      "kind": "normalization",
      "path_pattern": "model.text_model.encoder.layer.{i}.attention.output.LayerNorm"
    },
    {
      "attributes": {
        "head_dim": 96,
        "n_heads": 8,
        "n_kv_heads": 8,
        "rope": false,
        "variant": "MHA"
      },
      "children": [
        "encoder_layer.crossattention.self",
        "encoder_layer.crossattention.output"
      ],
      "class_name": "BlipTextAttention",
      "id": "encoder_layer.crossattention",
      "kind": "cross_attention",
      "path_pattern": "model.text_model.encoder.layer.{i}.crossattention"
    },
    {
      "attributes": {
        "head_dim": 96,
        "n_heads": 8,
        "n_kv_heads": 8,
        "rope": false,
        "variant": "MHA"
      },
      "class_name": "BlipTextSelfAttention",
      "id": "encoder_layer.crossattention.self",
      "kind": "cross_attention",
      "path_pattern": "model.text_model.encoder.layer.{i}.crossattention.self"
    },
    {
      "attributes": {
        "head_dim": 96,
        "n_heads": 8,
        "n_kv_heads": 8,
        "rope": false,
        "variant": "MHA"
      },
      "children": [
        "encoder_layer.crossattention.output.LayerNorm"
      ],
      "class_name": "BlipTextSelfOutput",
      "id": "encoder_layer.crossattention.output",
      "kind": "cross_attention",
      "path_pattern": "model.text_model.encoder.layer.{i}.crossattention.output"
    },
    {
      "attributes": {
        "head_dim": 96,
        "n_heads": 8,
        "n_kv_heads": 8,
        "rope": false,
        "variant": "MHA"
      },
      "class_name": "LayerNorm",
      "id": "encoder_layer.crossattention.output.LayerNorm",
      "kind": "cross_attention",
      "path_pattern": "model.text_model.encoder.layer.{i}.crossattention.output.LayerNorm"
    },
    {
      "attributes": {
        "activation": "gelu",
        "hidden_size": 768,
        "intermediate_size": 3072
      },
      "children": [
        "encoder_layer.intermediate.dense"
      ],
      "class_name": "BlipTextIntermediate",
      "id": "encoder_layer.intermediate",
      "kind": "feed_forward",
      "path_pattern": "model.text_model.encoder.layer.{i}.intermediate"
    },
    {
      "attributes": {
        "in_features": "config.hidden_size",
        "out_features": "config.intermediate_size"
      },
      "class_name": "Linear",
      "id": "encoder_layer.intermediate.dense",
      "kind": "projection",
      "path_pattern": "model.text_model.encoder.layer.{i}.intermediate.dense"
    },
    {
      "attributes": {
        "activation": "gelu",
        "hidden_size": 768,
        "intermediate_size": 3072
      },
      "children": [
        "encoder_layer.output.dense",
        "encoder_layer.output.LayerNorm"
      ],
      "class_name": "BlipTextOutput",
      "id": "encoder_layer.output",
      "kind": "feed_forward",
      "path_pattern": "model.text_model.encoder.layer.{i}.output"
    },
    {
      "attributes": {
        "in_features": "config.intermediate_size",
        "out_features": "config.hidden_size"
      },
      "class_name": "Linear",
      "id": "encoder_layer.output.dense",
      "kind": "projection",
      "path_pattern": "model.text_model.encoder.layer.{i}.output.dense"
    },
    {
      "attributes": {
        "norm_type": "layer"
      },
      "class_name": "LayerNorm",
      "id": "encoder_layer.output.LayerNorm",
      "kind": "normalization",
      "path_pattern": "model.text_model.encoder.layer.{i}.output.LayerNorm"
    },
    {
      "children": [
        "encoder_layer_2.self_attn",
        "encoder_layer_2.layer_norm1",
        "encoder_layer_2.mlp",
        "encoder_layer_2.layer_norm2"
      ],
      "class_name": "BlipEncoderLayer",
      "id": "encoder_layer_2",
      "kind": "transformer_block",
      "path_pattern": "model.vision_model.encoder.layers.{i}"
    },
    {
      "attributes": {
        "head_dim": 64,
        "n_heads": 12,
        "n_kv_heads": 12,
        "rope": false,
        "variant": "MHA"
      },
      "children": [
        "encoder_layer_2.self_attn.qkv",
        "encoder_layer_2.self_attn.projection"
      ],
      "class_name": "BlipAttention",
      "id": "encoder_layer_2.self_attn",
      "kind": "attention",
      "path_pattern": "model.vision_model.encoder.layers.{i}.self_attn"
    },
    {
      "attributes": {
        "in_features": "config.hidden_size",
        "out_features": 2304
      },
      "class_name": "Linear",
      "id": "encoder_layer_2.self_attn.qkv",
      "kind": "projection",
      "path_pattern": "model.vision_model.encoder.layers.{i}.self_attn.qkv"
    },
    {
      "attributes": {
        "in_features": "config.hidden_size",
        "out_features": "config.hidden_size"
      },
      "class_name": "Linear",
      "id": "encoder_layer_2.self_attn.projection",
      "kind": "projection",
      "path_pattern": "model.vision_model.encoder.layers.{i}.self_attn.projection"
    },
    {
      "attributes": {
        "norm_type": "layer"
      },
      "class_name": "LayerNorm",
      "id": "encoder_layer_2.layer_norm1",
      "kind": "normalization",
      "path_pattern": "model.vision_model.encoder.layers.{i}.layer_norm1"
    },
    {
      "attributes": {
        "activation": "gelu",
        "hidden_size": 768,
        "intermediate_size": 3072
      },
      "children": [
        "encoder_layer_2.mlp.fc1",
        "encoder_layer_2.mlp.fc2"
      ],
      "class_name": "BlipMLP",
      "id": "encoder_layer_2.mlp",
      "kind": "feed_forward",
      "path_pattern": "model.vision_model.encoder.layers.{i}.mlp"
    },
    {
      "attributes": {
        "in_features": "config.hidden_size",
        "out_features": "config.intermediate_size"
      },
      "class_name": "Linear",
      "id": "encoder_layer_2.mlp.fc1",
      "kind": "projection",
      "path_pattern": "model.vision_model.encoder.layers.{i}.mlp.fc1"
    },
    {
      "attributes": {
        "in_features": "config.intermediate_size",
        "out_features": "config.hidden_size"
      },
      "class_name": "Linear",
      "id": "encoder_layer_2.mlp.fc2",
      "kind": "projection",
      "path_pattern": "model.vision_model.encoder.layers.{i}.mlp.fc2"
    },
    {
      "attributes": {
        "norm_type": "layer"
      },
      "class_name": "LayerNorm",
      "id": "encoder_layer_2.layer_norm2",
      "kind": "normalization",
      "path_pattern": "model.vision_model.encoder.layers.{i}.layer_norm2"
    }
  ]
}