Self_Explaining_Structures_Improve_NLP_Models
Self_Explaining_Structures_Improve_NLP_Models copied to clipboard
Published
20 hours ago •
ShannonAI
ShannonAI
Baseline results without Self-Explaining
In your paper, you reported the following results on SST-5 for RoBERTa without Self-Explaining as a baseline:
| Model | Accuracy |
|---|---|
| RoBERTa-base | 56.4% |
| RoBERTa-large | 57.9% |
The original paper by Liu et al. (2019b) does not list any results for SST-5, so I'm assuming you obtained these results yourself. Could you share how you did that? Did you fine-tune these baselines on the SST-5 dataset, or are these the performances right out of the box? Many thanks in advance.
I have faced the same problem. I use the huggingface official script to finetune Roberta-base. However, I can hardly achieve such results. Have you reproduced the baseline and would you mind sharing the configuration?
