metadata

model_creators:
  - >-
    Leonardo Zilio, Hadeel Saadany, Prashant Sharma, Diptesh Kanojia, Constantin
    Orasan
license: mit
tags:
  - generated_from_trainer
datasets:
  - surrey-nlp/PLOD-unfiltered
metrics:
  - precision
  - recall
  - f1
  - accuracy
language:
  - en
widget:
  - text: >-
      Light dissolved inorganic carbon (DIC) resulting from the oxidation of
      hydrocarbons.
  - text: >-
      RAFs are plotted for a selection of neurons in the dorsal zone (DZ) of
      auditory cortex in Figure 1.
  - text: >-
      Images were acquired using a GE 3.0T MRI scanner with an upgrade for
      echo-planar imaging (EPI).
model-index:
  - name: roberta-large-finetuned-ner
    results:
      - task:
          name: Token Classification
          type: token-classification
        dataset:
          name: surrey-nlp/PLOD-unfiltered
          type: token-classification
          args: PLODunfiltered
        metrics:
          - name: Precision
            type: precision
            value: 0.9662545190541101
          - name: Recall
            type: recall
            value: 0.9627013733169376
          - name: F1
            type: f1
            value: 0.9644746737300262
          - name: Accuracy
            type: accuracy
            value: 0.9607518572002093

roberta-large-finetuned-ner

This model is a fine-tuned version of roberta-large on the PLOD-unfiltered dataset. It achieves the following results on the evaluation set:

Loss: 0.1393
Precision: 0.9663
Recall: 0.9627
F1: 0.9645
Accuracy: 0.9608

Model description

RoBERTa is a transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of publicly available data) with an automatic process to generate inputs and labels from those texts.

More precisely, it was pretrained with the Masked language modeling (MLM) objective. Taking a sentence, the model randomly masks 15% of the words in the input then run the entire masked sentence through the model and has to predict the masked words. This is different from traditional recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the sentence.

This way, the model learns an inner representation of the English language that can then be used to extract features useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard classifier using the features produced by the BERT model as inputs.

Intended uses & limitations

More information needed

Training and evaluation data

The model is fine-tuned using PLOD-Unfiltered dataset. This dataset is used for training and evaluating the model. The PLOD Dataset is published at LREC 2022. The dataset can help build sequence labeling models for the task of Abbreviation Detection.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 8
eval_batch_size: 4
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 6

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
0.1281	1.0	14233	0.1300	0.9557	0.9436	0.9496	0.9457
0.1056	2.0	28466	0.1076	0.9620	0.9552	0.9586	0.9545
0.0904	3.0	42699	0.1054	0.9655	0.9585	0.9620	0.9583
0.0743	4.0	56932	0.1145	0.9658	0.9602	0.9630	0.9593
0.0523	5.0	71165	0.1206	0.9664	0.9619	0.9641	0.9604
0.044	6.0	85398	0.1393	0.9663	0.9627	0.9645	0.9608

Framework versions

Transformers 4.18.0
Pytorch 1.10.1+cu111
Datasets 2.1.0
Tokenizers 0.12.1