rttl-ai/SentyBert

Model Details

Model Description: This model is a fine-tune checkpoint of bert-large-uncased, fine-tuned on SST-2. This model reaches an accuracy of 99.92 on the dev set.

Developed by: rttl-ai
Model Type: Text Classification
Language(s): English
License: Apache-2.0
Resources for more information:
The model was pre-trained with task-adaptive pre-training TAPT with an increased masking rate, no corruption strategy, and using WWM, following this paper
- fine-tuned on sst with subtrees
- fine-tuned on sst2

Downloads last month: 7

Datasets used to train rttl-ai/bert-large-uncased-sentiment

Paper for rttl-ai/bert-large-uncased-sentiment

Don't Stop Pretraining: Adapt Language Models to Domains and Tasks

Paper • 2004.10964 • Published Apr 23, 2020

Evaluation results

F1 Macro on sst2
validation set self-reported

0.999
Accuracy on sst2
validation set self-reported

0.999