Spaces:
Sleeping
Sleeping
| title: Fine-Tuned BERT Model | |
| emoji: 🌍 | |
| colorFrom: blue | |
| colorTo: purple | |
| sdk: docker | |
| pinned: true | |
| # Fine-Tuned BERT Model for Climate Disinformation Classification | |
| ## Model Description | |
| This is a fine-tuned BERT model trained for the Frugal AI Challenge 2024. The model has been fine-tuned on the climate disinformation dataset to classify text inputs into 8 distinct categories related to climate disinformation. It leverages BERT's pretrained language understanding capabilities and has been optimized for accuracy in this domain. | |
| ### Intended Use | |
| - **Primary intended uses**: Classifying text inputs to detect specific claims of climate disinformation | |
| - **Primary intended users**: Researchers, developers, and participants in the Frugal AI Challenge | |
| - **Out-of-scope use cases**: Not recommended for tasks outside climate disinformation classification or production-level applications without further evaluation | |
| ## Training Data | |
| The model uses the QuotaClimat/frugalaichallenge-text-train dataset: | |
| - Size: ~6000 examples | |
| - Split: 80% train, 20% test | |
| - 8 categories of climate disinformation claims | |
| ### Labels | |
| 0. No relevant claim detected | |
| 1. Global warming is not happening | |
| 2. Not caused by humans | |
| 3. Not bad or beneficial | |
| 4. Solutions harmful/unnecessary | |
| 5. Science is unreliable | |
| 6. Proponents are biased | |
| 7. Fossil fuels are needed | |
| ## Performance | |
| ### Metrics | |
| - **Accuracy**: Achieved XX.X% on the test set (replace `XX.X%` with the actual accuracy from your evaluation) | |
| - **Environmental Impact**: | |
| - Carbon emissions tracked in gCO2eq | |
| - Energy consumption tracked in Wh | |
| ### Model Architecture | |
| This model fine-tunes the BERT base architecture (`bert-base-uncased`) for the climate disinformation task. The classifier head includes: | |
| - Dense layers | |
| - Dropout for regularization | |
| - Softmax activation for multi-class classification | |
| ## Environmental Impact | |
| Environmental impact is tracked using CodeCarbon, measuring: | |
| - Carbon emissions during inference and training | |
| - Energy consumption during inference and training | |
| This tracking aligns with the Frugal AI Challenge's commitment to promoting sustainable AI practices. | |
| ## Limitations | |
| - Fine-tuned specifically for climate disinformation; performance on other text classification tasks may degrade | |
| - Requires computational resources (e.g., GPU) for efficient inference | |
| - Predictions rely on the training dataset's representativeness; may struggle with unseen or out-of-distribution data | |
| ## Ethical Considerations | |
| - Dataset contains sensitive topics related to climate disinformation | |
| - Model performance depends on the quality of the dataset and annotation biases | |
| - Environmental impact during training and inference is disclosed to encourage awareness of AI's carbon footprint | |
| - Users must validate outputs before using in sensitive or high-stakes applications | |
| --- | |