Python 将模型配置从allennlp 0.9.0更新为1.3.0

我正在尝试使用Udify库对依赖项解析的预训练多语言BERT模型进行微调。这个库使用了allennlp==0.9.0，而我需要使用allennlp=1.3.0，并且我正在尝试为此更新代码。在对配置文件进行一些更改之后，我发现模型的forward（）方法中的输入有问题。也就是说，

tokens

被传递，在我的例子中，它是一个如下所示的字典：

tokens = {
            "bert": {
                "bert": tensor(...),
                "bert-offsets": tensor(...),
                "bert-type-ids": tensor(...),
                "mask": tensor(...)
            },
            "tokens": {
                "tokens": tensor(...)
            }
        }

错误发生在：

适用于allennlp 0.9.0的相应配置如下：

{
  "dataset_reader": {
    "lazy": false,
    "token_indexers": {
      "tokens": {
        "type": "single_id",
        "lowercase_tokens": true
      },
      "bert": {
        "type": "udify-bert-pretrained",
        "pretrained_model": "config/archive/bert-base-multilingual-cased/vocab.txt",
        "do_lowercase": false,
        "use_starting_offsets": true
      }
    }
  },
  "train_data_path": "data/ud-treebanks-v2.3/UD_English-EWT/en_ewt-ud-train.conllu",
  "validation_data_path": "data/ud-treebanks-v2.3/UD_English-EWT/en_ewt-ud-dev.conllu",
  "test_data_path": "data/ud-treebanks-v2.3/UD_English-EWT/en_ewt-ud-test.conllu",
  "vocabulary": {
    "directory_path": "data/vocab/english_only_expmix4/vocabulary"
  },
  "model": {
    "word_dropout": 0.2,
    "mix_embedding": 12,
    "layer_dropout": 0.1,
    "tasks": ["deps"],
    "text_field_embedder": {
      "type": "udify_embedder",
      "dropout": 0.5,
      "allow_unmatched_keys": true,
      "embedder_to_indexer_map": {
        "bert": ["bert", "bert-offsets"]
      },
      "token_embedders": {
        "bert": {
          "type": "udify-bert-pretrained",
          "pretrained_model": "bert-base-multilingual-cased",
          "requires_grad": true,
          "dropout": 0.15,
          "layer_dropout": 0.1,
          "combine_layers": "all"
        }
      }
    },
    "encoder": {
      "type": "pass_through",
      "input_dim": 768
    },
    "decoders": {
      "upos": {
        "encoder": {
          "type": "pass_through",
          "input_dim": 768
        }
      },
      "feats": {
        "encoder": {
          "type": "pass_through",
          "input_dim": 768
        },
        "adaptive": true
      },
      "lemmas": {
        "encoder": {
          "type": "pass_through",
          "input_dim": 768
        },
        "adaptive": true
      },
      "deps": {
        "tag_representation_dim": 256,
        "arc_representation_dim": 768,
        "encoder": {
          "type": "pass_through",
          "input_dim": 768
        }
      }
    }
  },
  "iterator": {
    "batch_size": 16
  },
  "trainer": {
    "num_epochs": 5,
    "patience": 40,
    "num_serialized_models_to_keep": 1,
    "should_log_learning_rate": true,
    "summary_interval": 100,
    "optimizer": {
      "type": "bert_adam",
      "b1": 0.9,
      "b2": 0.99,
      "weight_decay": 0.01,
      "lr": 1e-3,
      "parameter_groups": [
        [["^text_field_embedder.*.bert_model.embeddings",
          "^text_field_embedder.*.bert_model.encoder"], {}],
        [["^text_field_embedder.*._scalar_mix",
          "^text_field_embedder.*.pooler",
          "^scalar_mix",
          "^decoders",
          "^shared_encoder"], {}]
      ]
    },
    "learning_rate_scheduler": {
      "type": "ulmfit_sqrt",
      "model_size": 1,
      "warmup_steps": 392,
      "start_step": 392,
      "factor": 5.0,
      "gradual_unfreezing": true,
      "discriminative_fine_tuning": true,
      "decay_factor": 0.04
    }
  },
  "udify_replace": [
    "dataset_reader.token_indexers",
    "model.text_field_embedder",
    "model.encoder",
    "model.decoders.xpos",
    "model.decoders.deps.encoder",
    "model.decoders.upos.encoder",
    "model.decoders.feats.encoder",
    "model.decoders.lemmas.encoder",
    "trainer.learning_rate_scheduler",
    "trainer.optimizer"
  ]
}

---

嗨，你找到解决办法了吗？更改配置后，我出现了此错误：“

返回火炬。嵌入（重量、输入、填充idx、比例、梯度、频率、稀疏）索引器：在self中索引超出范围

”

{
  "dataset_reader": {
    "lazy": false,
    "token_indexers": {
      "tokens": {
        "type": "single_id",
        "lowercase_tokens": true
      },
      "bert": {
        "type": "udify-bert-pretrained",
        "pretrained_model": "config/archive/bert-base-multilingual-cased/vocab.txt",
        "do_lowercase": false,
        "use_starting_offsets": true
      }
    }
  },
  "train_data_path": "data/ud-treebanks-v2.3/UD_English-EWT/en_ewt-ud-train.conllu",
  "validation_data_path": "data/ud-treebanks-v2.3/UD_English-EWT/en_ewt-ud-dev.conllu",
  "test_data_path": "data/ud-treebanks-v2.3/UD_English-EWT/en_ewt-ud-test.conllu",
  "vocabulary": {
    "type": "from_files",
    "directory": "data/concat-exp-mix/vocab/concat-exp-mix/vocabulary/"
  },
  "model": {
    "word_dropout": 0.2,
    "mix_embedding": 12,
    "layer_dropout": 0.1,
    "tasks": ["deps"],
    "pretrained_model": "bert-base-multilingual-cased",
    "text_field_embedder": {
      "type": "udify_embedder",
      "dropout": 0.5,
      "allow_unmatched_keys": true,
      "embedder_to_indexer_map": {
        "bert": ["bert", "bert-offsets"]
      },
      "token_embedders": {
        "bert": {
          "type": "udify-bert-pretrained",
          "pretrained_model": "bert-base-multilingual-cased",
          "requires_grad": true,
          "dropout": 0.15,
          "layer_dropout": 0.1,
          "combine_layers": "all"
        }
      }
    },
    "encoder": {
      "type": "pass_through",
      "input_dim": 768
    },
    "decoders": {
      "upos": {
        "encoder": {
          "type": "pass_through",
          "input_dim": 768
        }
      },
      "feats": {
        "encoder": {
          "type": "pass_through",
          "input_dim": 768
        },
        "adaptive": true
      },
      "lemmas": {
        "encoder": {
          "type": "pass_through",
          "input_dim": 768
        },
        "adaptive": true
      },
      "deps": {
        "tag_representation_dim": 256,
        "arc_representation_dim": 768,
        "encoder": {
          "type": "pass_through",
          "input_dim": 768
        }
      }
    }
  },
  "data_loader": {
    "batch_sampler":{
      "batch_size": 16
    }

  },
  "trainer": {
    "num_epochs": 5,
    "patience": 40,
    "optimizer": {
      "type": "adamw",
      "betas": [0.9, 0.99],
      "weight_decay": 0.01,
      "lr": 1e-3,
      "parameter_groups": [
        [["^text_field_embedder.*.bert_model.embeddings",
          "^text_field_embedder.*.bert_model.encoder"], {}],
        [["^text_field_embedder.*._scalar_mix",
          "^text_field_embedder.*.pooler",
          "^scalar_mix",
          "^decoders",
          "^shared_encoder"], {}]
      ]
    },
    "learning_rate_scheduler": {
      "type": "ulmfit_sqrt",
      "model_size": 1,
      "warmup_steps": 392,
      "start_step": 392,
      "factor": 5.0,
      "gradual_unfreezing": true,
      "discriminative_fine_tuning": true,
      "decay_factor": 0.04
    }
  },
  "udify_replace": [
    "dataset_reader.token_indexers",
    "model.text_field_embedder",
    "model.encoder",
    "model.decoders.xpos",
    "model.decoders.deps.encoder",
    "model.decoders.upos.encoder",
    "model.decoders.feats.encoder",
    "model.decoders.lemmas.encoder",
    "trainer.learning_rate_scheduler",
    "trainer.optimizer"
  ]
}

{
  "dataset_reader": {
    "lazy": false,
    "token_indexers": {
      "tokens": {
        "type": "single_id",
        "lowercase_tokens": true
      },
      "bert": {
        "type": "udify-bert-pretrained",
        "pretrained_model": "config/archive/bert-base-multilingual-cased/vocab.txt",
        "do_lowercase": false,
        "use_starting_offsets": true
      }
    }
  },
  "train_data_path": "data/ud-treebanks-v2.3/UD_English-EWT/en_ewt-ud-train.conllu",
  "validation_data_path": "data/ud-treebanks-v2.3/UD_English-EWT/en_ewt-ud-dev.conllu",
  "test_data_path": "data/ud-treebanks-v2.3/UD_English-EWT/en_ewt-ud-test.conllu",
  "vocabulary": {
    "directory_path": "data/vocab/english_only_expmix4/vocabulary"
  },
  "model": {
    "word_dropout": 0.2,
    "mix_embedding": 12,
    "layer_dropout": 0.1,
    "tasks": ["deps"],
    "text_field_embedder": {
      "type": "udify_embedder",
      "dropout": 0.5,
      "allow_unmatched_keys": true,
      "embedder_to_indexer_map": {
        "bert": ["bert", "bert-offsets"]
      },
      "token_embedders": {
        "bert": {
          "type": "udify-bert-pretrained",
          "pretrained_model": "bert-base-multilingual-cased",
          "requires_grad": true,
          "dropout": 0.15,
          "layer_dropout": 0.1,
          "combine_layers": "all"
        }
      }
    },
    "encoder": {
      "type": "pass_through",
      "input_dim": 768
    },
    "decoders": {
      "upos": {
        "encoder": {
          "type": "pass_through",
          "input_dim": 768
        }
      },
      "feats": {
        "encoder": {
          "type": "pass_through",
          "input_dim": 768
        },
        "adaptive": true
      },
      "lemmas": {
        "encoder": {
          "type": "pass_through",
          "input_dim": 768
        },
        "adaptive": true
      },
      "deps": {
        "tag_representation_dim": 256,
        "arc_representation_dim": 768,
        "encoder": {
          "type": "pass_through",
          "input_dim": 768
        }
      }
    }
  },
  "iterator": {
    "batch_size": 16
  },
  "trainer": {
    "num_epochs": 5,
    "patience": 40,
    "num_serialized_models_to_keep": 1,
    "should_log_learning_rate": true,
    "summary_interval": 100,
    "optimizer": {
      "type": "bert_adam",
      "b1": 0.9,
      "b2": 0.99,
      "weight_decay": 0.01,
      "lr": 1e-3,
      "parameter_groups": [
        [["^text_field_embedder.*.bert_model.embeddings",
          "^text_field_embedder.*.bert_model.encoder"], {}],
        [["^text_field_embedder.*._scalar_mix",
          "^text_field_embedder.*.pooler",
          "^scalar_mix",
          "^decoders",
          "^shared_encoder"], {}]
      ]
    },
    "learning_rate_scheduler": {
      "type": "ulmfit_sqrt",
      "model_size": 1,
      "warmup_steps": 392,
      "start_step": 392,
      "factor": 5.0,
      "gradual_unfreezing": true,
      "discriminative_fine_tuning": true,
      "decay_factor": 0.04
    }
  },
  "udify_replace": [
    "dataset_reader.token_indexers",
    "model.text_field_embedder",
    "model.encoder",
    "model.decoders.xpos",
    "model.decoders.deps.encoder",
    "model.decoders.upos.encoder",
    "model.decoders.feats.encoder",
    "model.decoders.lemmas.encoder",
    "trainer.learning_rate_scheduler",
    "trainer.optimizer"
  ]
}