scibert-scivocab-cased_pub_section

• original model file name: textclassifer_scibert_scivocab_cased_pubmed_20k
• This is a fine-tuned checkpoint of allenai/scibert_scivocab_cased for document section text classification
• possible document section classes are:BACKGROUND, CONCLUSIONS, METHODS, OBJECTIVE, RESULTS,

metadata

training_metrics

• date_run: Apr-26-2022_t-13

• huggingface_tag: allenai/scibert_scivocab_cased

• test_set: [{'test_accuracy': 0.8313589096069336, 'test_matthewscorrcoef': 0.7736952900886536, 'test_f1score': 0.8317078948020935, 'test_cross_entropy': 0.5242752432823181}]

training_parameters

• NUM_EPOCHS: 12

• BATCH_SIZE: 32

• MAX_INPUT_LENGTH: 256

• TRAIN_FP16: True

• TRAIN_STRATEGY: freeze

• LR_SCHEDULE: reducelronplateau

• LR_INITIAL: 0.001

• WEIGHT_DECAY: 0.05

• UNFREEZE_EPOCH: 4

• hf_tag: allenai/scibert_scivocab_cased

• lowercased_input: False

• input_text_colname: description

• target_cls_colname: target

• num_classes: 5

• model_shortname: scibert_scivocab_cased