test_config.yaml

# Basic config file
# comment for empty args

VOCABULARY:
    huggingface_model: "Enoch/graphcodebert-py"

READER:
    huggingface_model: "Enoch/graphcodebert-py"
    snippet_splitter: "\n$$$\n"
    label_splitter: " $x$ "
    multi_labels: " ; "
    part_graph: [256, 256]
    # kwargs_tokenizer:
    # kwargs_indexer:

MODEL:
    labels:
        - "miss_return"
        - "print_return"
        - "bad_division"
        - "bad_variable"
        - "bad_loop"
        - "bad_range"
        - "bad_assign"
        - "bad_list"
        - "bad_file"
        - "miss_try"
        - "miss_parenthesis"
        - "hardcoded_arg"
        - "overwrite_not_increment"
        - "miss_loop"
        - "failed"
        - "correct"
    huggingface_model: "Enoch/graphcodebert-py"
    kwargs_embedder:
        trainable: True
    embedding_size: 768
    encoder:
        name: "bert_pooler"
        arg: ["Enoch/graphcodebert-py"]
        # kwargs:
    classification_head:
        name: "mult_dense"
        arg: [ 1, 768, 18 ]
        kwargs:
            activation:
                name: "leaky_relu"
                # arg:
                # kwargs:
            norm: True
   # accuracy:
        # name: "categorical_accuracy"
        # arg:
        # kwargs:
    loss:
        name: "multilabel_soft_margin_loss"
    multi_label: True
    debug: False

TRAINER:
    validation_metric: "-loss"
    learning_rate: 1.e-5

INTERPRETER:
    captum: True
    kwargs:
        interpreter_name: "LayerIntegratedGradients"
        layer: "bert_interpretable_layer"
        attribute_kwargs:
            n_steps: 10
            internal_batch_size: 1

CONFIG:
    predict: True
    no_loop: True
    no_eval: True
    gui: False
    model: "Enoch/graphcodebert-py"
    training: "demo/train.txt"
    validation: "demo/validation.txt"
    evaluation: "demo/test.txt"
    serialization_dir: "demo"
    no_loop_weight_file: "model_weights.th"
    batch_size: 8
    loops: 10
    device: "cuda"