{
  "architecture": {
    "attention_variant": "MHA",
    "family": "masked_lm",
    "is_moe": false,
    "mixer": "attention",
    "positional": "learned",
    "tie_word_embeddings": true,
    "view": "encoder"
  },
  "capabilities": {
    "attention_backends": [
      "eager"
    ],
    "attention_patterns": [
      "bidirectional"
    ],
    "attention_schedule": null,
    "task_heads": [
      "masked_lm",
      "pretraining",
      "question_answering",
      "sequence_classification",
      "text_encoding",
      "token_classification"
    ],
    "tensor_parallel": false
  },
  "components": [
    {
      "children": [
        "embeddings",
        "encoder"
      ],
      "class_name": "DebertaModel",
      "id": "model",
      "kind": "model",
      "path_pattern": "model"
    },
    {
      "attributes": {
        "embedding_dim": "config.hidden_size",
        "num_embeddings": "config.vocab_size"
      },
      "children": [
        "embeddings.word_embeddings",
        "embeddings.position_embeddings",
        "embeddings.LayerNorm"
      ],
      "class_name": "DebertaEmbeddings",
      "id": "embeddings",
      "kind": "embedding",
      "path_pattern": "model.embeddings"
    },
    {
      "attributes": {
        "embedding_dim": "config.hidden_size",
        "num_embeddings": "config.vocab_size",
        "tied_lm_head": true
      },
      "class_name": "Embedding",
      "id": "embeddings.word_embeddings",
      "kind": "embedding",
      "path_pattern": "model.embeddings.word_embeddings"
    },
    {
      "attributes": {
        "max_position_embeddings": "config.max_position_embeddings",
        "scheme": "learned"
      },
      "class_name": "Embedding",
      "id": "embeddings.position_embeddings",
      "kind": "position",
      "path_pattern": "model.embeddings.position_embeddings"
    },
    {
      "attributes": {
        "norm_type": "layer"
      },
      "class_name": "DebertaLayerNorm",
      "id": "embeddings.LayerNorm",
      "kind": "normalization",
      "path_pattern": "model.embeddings.LayerNorm"
    },
    {
      "children": [
        "encoder_layers"
      ],
      "class_name": "DebertaEncoder",
      "id": "encoder",
      "kind": "encoder",
      "path_pattern": "model.encoder"
    }
  ],
  "config": {
    "class_name": "DebertaConfig",
    "model_type": "deberta",
    "module": "transformers.models.deberta.configuration_deberta",
    "referenced_fields": {
      "num_hidden_layers": 12
    },
    "salient_fields": {
      "hidden_act": "gelu",
      "hidden_size": 768,
      "intermediate_size": 3072,
      "is_encoder_decoder": false,
      "max_position_embeddings": 512,
      "num_attention_heads": 12,
      "num_hidden_layers": 12,
      "tie_word_embeddings": true,
      "vocab_size": 50265
    }
  },
  "edges": [
    {
      "kind": "data",
      "source": "embeddings",
      "target": "encoder"
    },
    {
      "kind": "data",
      "source": "embeddings.word_embeddings",
      "target": "embeddings.LayerNorm"
    },
    {
      "kind": "data",
      "source": "encoder_layer.attention",
      "target": "encoder_layer.intermediate"
    },
    {
      "kind": "data",
      "source": "encoder_layer.intermediate",
      "target": "encoder_layer.output"
    },
    {
      "kind": "mask",
      "source": "input:attention_mask",
      "target": "encoder_layer.attention"
    },
    {
      "kind": "mask",
      "source": "input:attention_mask",
      "target": "encoder_layer.attention.self"
    },
    {
      "kind": "residual",
      "source": "encoder_layer",
      "target": "encoder_layer.attention"
    },
    {
      "kind": "residual",
      "source": "encoder_layer",
      "target": "encoder_layer.intermediate"
    },
    {
      "kind": "position",
      "source": "embeddings.position_embeddings",
      "target": "embeddings"
    },
    {
      "kind": "data",
      "provenance": "intra_module",
      "source": "encoder_layer.attention.self",
      "target": "encoder_layer.attention.self.in_proj"
    },
    {
      "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"
    }
  ],
  "extends": null,
  "model_type": "deberta",
  "provenance": {
    "config_class": "DebertaConfig",
    "config_module": "transformers.models.deberta.configuration_deberta",
    "model_class": "DebertaModel",
    "model_module": "transformers.models.deberta.modeling_deberta"
  },
  "repeats": [
    {
      "body": "encoder_layer",
      "container_path_pattern": "model.encoder.layer",
      "count": 12,
      "count_expr": "config.num_hidden_layers",
      "count_source": "config",
      "id": "encoder_layers",
      "index_symbol": "i",
      "item_path_pattern": "model.encoder.layer.{i}",
      "kind": "symbolic_repeat",
      "provenance": {
        "class_name": "DebertaLayer",
        "container_path_pattern": "model.encoder.layer",
        "item_path_pattern": "model.encoder.layer.{i}",
        "source": "module_tree_repeat_collapse"
      },
      "repeated_class_name": "DebertaLayer"
    }
  ],
  "schema_version": "architecture-template-v0",
  "templates": [
    {
      "children": [
        "encoder_layer.attention",
        "encoder_layer.intermediate",
        "encoder_layer.output"
      ],
      "class_name": "DebertaLayer",
      "id": "encoder_layer",
      "kind": "transformer_block",
      "path_pattern": "model.encoder.layer.{i}"
    },
    {
      "attributes": {
        "head_dim": 64,
        "n_heads": 12,
        "n_kv_heads": 12,
        "pattern": "bidirectional",
        "rope": false,
        "variant": "MHA"
      },
      "children": [
        "encoder_layer.attention.self",
        "encoder_layer.attention.output"
      ],
      "class_name": "DebertaAttention",
      "id": "encoder_layer.attention",
      "kind": "attention",
      "path_pattern": "model.encoder.layer.{i}.attention"
    },
    {
      "attributes": {
        "head_dim": 64,
        "n_heads": 12,
        "n_kv_heads": 12,
        "pattern": "bidirectional",
        "rope": false,
        "variant": "MHA"
      },
      "children": [
        "encoder_layer.attention.self.in_proj"
      ],
      "class_name": "DisentangledSelfAttention",
      "id": "encoder_layer.attention.self",
      "kind": "attention",
      "path_pattern": "model.encoder.layer.{i}.attention.self"
    },
    {
      "attributes": {
        "in_features": "config.hidden_size",
        "out_features": 2304
      },
      "class_name": "Linear",
      "id": "encoder_layer.attention.self.in_proj",
      "kind": "projection",
      "path_pattern": "model.encoder.layer.{i}.attention.self.in_proj"
    },
    {
      "children": [
        "encoder_layer.attention.output.LayerNorm"
      ],
      "class_name": "DebertaSelfOutput",
      "id": "encoder_layer.attention.output",
      "kind": "module",
      "path_pattern": "model.encoder.layer.{i}.attention.output"
    },
    {
      "attributes": {
        "norm_type": "layer"
      },
      "class_name": "DebertaLayerNorm",
      "id": "encoder_layer.attention.output.LayerNorm",
      "kind": "normalization",
      "path_pattern": "model.encoder.layer.{i}.attention.output.LayerNorm"
    },
    {
      "attributes": {
        "activation": "gelu",
        "hidden_size": 768,
        "intermediate_size": 3072
      },
      "children": [
        "encoder_layer.intermediate.dense"
      ],
      "class_name": "DebertaIntermediate",
      "id": "encoder_layer.intermediate",
      "kind": "feed_forward",
      "path_pattern": "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.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": "DebertaOutput",
      "id": "encoder_layer.output",
      "kind": "feed_forward",
      "path_pattern": "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.encoder.layer.{i}.output.dense"
    },
    {
      "attributes": {
        "norm_type": "layer"
      },
      "class_name": "DebertaLayerNorm",
      "id": "encoder_layer.output.LayerNorm",
      "kind": "normalization",
      "path_pattern": "model.encoder.layer.{i}.output.LayerNorm"
    }
  ]
}