metadata
tags:
- generated_from_keras_callback
model-index:
- name: GeoBERT
results: []
GeoBERT
GeoBERT is a NER model that was fine-tuned from SciBERT on the Geoscientific Corpus dataset. The model was trained on the Labeled Geoscientific Corpus dataset (~1 million sentences).
Intended uses
The NER product in this model has a goal to identify four main semantic types or categories related to Geosciences.
- GeoPetro for any entities that belong to all terms in Geosciences
- GeoMeth for all tools or methods associated with Geosciences
- GeoLoc to identify geological locations
- GeoTime for identifying the geological time scale entities
Training hyperparameters
The following hyperparameters were used during training:
- optimizer: {'name': 'AdamWeightDecay', 'learning_rate': {'class_name': 'PolynomialDecay', 'config': {'initial_learning_rate': 2e-05, 'decay_steps': 14000, 'end_learning_rate': 0.0, 'power': 1.0, 'cycle': False, 'name': None}}, 'decay': 0.0, 'beta_1': 0.9, 'beta_2': 0.999, 'epsilon': 1e-08, 'amsgrad': False, 'weight_decay_rate': 0.01}
- training_precision: mixed_float16
Framework versions
- Transformers 4.22.1
- TensorFlow 2.10.0
- Datasets 2.4.0
- Tokenizers 0.12.1
Model performances (metric: seqeval)
| entity | precision | recall | f1 |
|---|---|---|---|
| GeoLoc | 0.9727 | 0.9591 | 0.9658 |
| GeoMeth | 0.9433 | 0.9447 | 0.9445 |
| GeoPetro | 0.9767 | 0.9745 | 0.9756 |
| GeoTime | 0.9695 | 0.9666 | 0.9680 |
How to use GeoBERT with HuggingFace
Load GeoBERT and its sub-word tokenizer :
from transformers import AutoTokenizer, AutoModelForTokenClassification
tokenizer = AutoTokenizer.from_pretrained("botryan96/GeoBERT")
model = AutoModelForTokenClassification.from_pretrained("botryan96/GeoBERT")