metadata
language: en
license: apache-2.0
Using the DistilRoBERTa model as starting point, the ClimateBERT Language Model is additionally pretrained on a text corpus comprising climate-related research paper abstracts, corporate and general news and reports from companies. The underlying methodology can be found in our language model research paper.
Climate performance model card
| distilroberta-base-climate-f | |
|---|---|
| 1. Is the resulting model publicly available? | Yes |
| 2. How much time does the training of the final model take? | 8 hours |
| 3. How much time did all experiments take (incl. hyperparameter search)? | 288 hours |
| 4. What was the energy consumption (GPU/CPU)? | 0.7 kW |
| 5. At which geo location were the computations performed? | Germany |
| 6. What was the energy mix at the geo location? | 470 gCO2eq/kWh |
| 7. How much CO2eq was emitted to train the final model? | 2.63 kg |
| 8. How much CO2eq was emitted for all experiments? | 94.75 kg |
| 9. What is the average CO2eq emission for the inference of one sample? | 0.62 mg |
| 10. Which positive environmental impact can be expected from this work? | This work can be categorized as a building block tools following Jin et al (2021). It supports the training of NLP models in the field of climate change and, thereby, have a positive environmental impact in the future. |
| 11. Comments | Block pruning could decrease CO2eq emissions |
BibTeX entry and citation info
@article{wkbl2021,
title={ClimateBERT: A Pretrained Language Model for Climate-Related Text},
author={Webersinke, Nicolas and Kraus, Mathias and Bingler, Julia and Leippold, Markus},
journal={arXiv preprint arXiv:2110.12010},
year={2021}
}