_data/architectures/bert.json

{
    "model_type": "bert",
    "architectures": [
        "BertModel"
    ],
    "pre_weights": [
        {
            "name": "embeddings.position_embeddings.weight",
            "aliases": [
                "bert.embeddings.position_embeddings.weight"
            ]
        },
        {
            "name": "embeddings.token_type_embeddings.weight",
            "aliases": [
                "bert.embeddings.token_type_embeddings.weight"
            ]
        },
        {
            "name": "embeddings.word_embeddings.weight",
            "is_embed": true,
            "aliases": [
                "bert.embeddings.word_embeddings.weight"
            ]
        },
        {
            "name": "embeddings.LayerNorm.bias",
            "aliases": [
                "embeddings.LayerNorm.beta",
                "bert.embeddings.LayerNorm.bias",
                "bert.embeddings.LayerNorm.beta"
            ]
        },
        {
            "name": "embeddings.LayerNorm.weight",
            "aliases": [
                "embeddings.LayerNorm.gamma",
                "bert.embeddings.LayerNorm.weight",
                "bert.embeddings.LayerNorm.gamma",
                "bert.embeddings.LayerNorm.weight"
            ]
        },
        {
            "name": "embeddings.position_ids",
            "optional": true,
            "force_dtype": "int64",
            "aliases": [
                "bert.embeddings.position_ids"
            ]
        }
    ],
    "post_weights": [
        {
            "name": "pooler.dense.weight",
            "aliases": [
                "bert.pooler.dense.weight"
            ]
        },
        {
            "name": "pooler.dense.bias",
            "aliases": [
                "bert.pooler.dense.bias"
            ]
        }
    ],
    "num_layers_config_key": "num_hidden_layers",
    "layer_templates": {
        "weights": [
            {
                "name": "encoder.layer.${layer_index}.attention.self.query.weight",
                "aliases": [
                    "bert.encoder.layer.${layer_index}.attention.self.query.weight"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.attention.self.query.bias",
                "aliases": [
                    "bert.encoder.layer.${layer_index}.attention.self.query.bias"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.attention.self.key.weight",
                "aliases": [
                    "bert.encoder.layer.${layer_index}.attention.self.key.weight"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.attention.self.key.bias",
                "aliases": [
                    "bert.encoder.layer.${layer_index}.attention.self.key.bias"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.attention.self.value.weight",
                "aliases": [
                    "bert.encoder.layer.${layer_index}.attention.self.value.weight"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.attention.self.value.bias",
                "aliases": [
                    "bert.encoder.layer.${layer_index}.attention.self.value.bias"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.attention.output.dense.weight",
                "aliases": [
                    "bert.encoder.layer.${layer_index}.attention.output.dense.weight"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.attention.output.dense.bias",
                "aliases": [
                    "bert.encoder.layer.${layer_index}.attention.output.dense.bias"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.attention.output.LayerNorm.bias",
                "aliases": [
                    "encoder.layer.${layer_index}.attention.output.LayerNorm.beta",
                    "bert.encoder.layer.${layer_index}.attention.output.LayerNorm.bias",
                    "bert.encoder.layer.${layer_index}.attention.output.LayerNorm.beta"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.attention.output.LayerNorm.weight",
                "aliases": [
                    "encoder.layer.${layer_index}.attention.output.LayerNorm.gamma",
                    "bert.encoder.layer.${layer_index}.attention.output.LayerNorm.weight",
                    "bert.encoder.layer.${layer_index}.attention.output.LayerNorm.gamma"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.intermediate.dense.weight",
                "aliases": [
                    "bert.encoder.layer.${layer_index}.intermediate.dense.weight"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.intermediate.dense.bias",
                "aliases": [
                    "bert.encoder.layer.${layer_index}.intermediate.dense.bias"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.output.dense.weight",
                "aliases": [
                    "bert.encoder.layer.${layer_index}.output.dense.weight"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.output.dense.bias",
                "aliases": [
                    "bert.encoder.layer.${layer_index}.output.dense.bias"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.output.LayerNorm.bias",
                "aliases": [
                    "encoder.layer.${layer_index}.output.LayerNorm.beta",
                    "bert.encoder.layer.${layer_index}.output.LayerNorm.bias",
                    "bert.encoder.layer.${layer_index}.output.LayerNorm.beta"
                ]
            },
            {
                "name": "encoder.layer.${layer_index}.output.LayerNorm.weight",
                "aliases": [
                    "encoder.layer.${layer_index}.output.LayerNorm.gamma",
                    "bert.encoder.layer.${layer_index}.output.LayerNorm.weight",
                    "bert.encoder.layer.${layer_index}.output.LayerNorm.gamma"
                ]
            }
        ]
    }
}