cglez/bert-dapt-civil_comments-uncased

Model Card: BERT DAPT Civil Comments

A domain-adapted BERT-base model, further pre-trained on the Civil Comments dataset text.

Model Details

Description

This model is based on the BERT base (uncased) architecture and was further pre-trained (domain-adapted) using the text in Civil Comments dataset, excluding its test split. Only the masked language modeling (MLM) objective was used during domain adaptation.

• Developed by: Cesar Gonzalez-Gutierrez
• Funded by: ERC
• Architecture: BERT-base
• Language: English
• License: Apache 2.0
• Base model: BERT base model (uncased)

Checkpoints

Intermediate checkpoints from the pre-training process are available and can be accessed using specific tags, which correspond to training epochs and steps:

Epoch	Step	Tags
1	17833	epoch-1	step-17833
2	35666	epoch-2	step-35666
3	53499	epoch-3	step-53499
5	89166	epoch-5	step-89166
10	178333	epoch-10	step-178333
15	267495	epoch-15	step-267495
20	356660	epoch-20	step-356660
25	445825	epoch-25	step-445825

To load a model from a specific intermediate checkpoint, use the revision parameter with the corresponding tag:

from transformers import AutoModelForMaskedLM

model = AutoModelForMaskedLM.from_pretrained("<model-name>", revision="<checkpoint-tag>")

Sources

• Paper: [Information pending]

Training Details

For more details on the training procedure, please refer to the base model's documentation: Training procedure.

Training Data

All texts from Civil Comments dataset, excluding the test partition.

Preprocessing

All markup and symbols were removed from the texts, including punctuation.

Training Hyperparameters

• Precision: fp16
• Batch size: 32
• Gradient accumulation steps: 3

Uses

For typical use cases and limitations, please refer to the base model's guidance: Inteded uses & limitations.

Bias, Risks, and Limitations

This model inherits potential risks and limitations from the base model. Refer to: Limitations and bias.

Environmental Impact

• Hardware Type: NVIDIA Tesla V100 PCIE 32GB
• Runtime: 44.5 h
• Cluster Provider: Artemisa
• Compute Region: EU
• Carbon Emitted: 8.28 kg CO2 eq.

Citation

BibTeX:

[More Information Needed]