SpanMarker with bert-base-cased on Acronym Identification
This is a SpanMarker model trained on the Acronym Identification dataset that can be used for Named Entity Recognition. This SpanMarker model uses bert-base-cased as the underlying encoder. See train.py for the training script.
Is your data not (always) capitalized correctly? Then consider using the uncased variant of this model instead for better performance: tomaarsen/span-marker-bert-base-uncased-acronyms.
Model Details
Model Description
- Model Type: SpanMarker
- Encoder: bert-base-cased
- Maximum Sequence Length: 256 tokens
- Maximum Entity Length: 8 words
- Training Dataset: Acronym Identification
- Language: en
- License: apache-2.0
Model Sources
- Repository: SpanMarker on GitHub
- Thesis: SpanMarker For Named Entity Recognition
Model Labels
| Label | Examples |
|---|---|
| long | "Conversational Question Answering", "controlled natural language", "successive convex approximation" |
| short | "SODA", "CNL", "CoQA" |
Evaluation
Metrics
| Label | Precision | Recall | F1 |
|---|---|---|---|
| all | 0.9422 | 0.9252 | 0.9336 |
| long | 0.9308 | 0.9013 | 0.9158 |
| short | 0.9479 | 0.9374 | 0.9426 |
Uses
Direct Use for Inference
from span_marker import SpanMarkerModel
# Download from the 🤗 Hub
model = SpanMarkerModel.from_pretrained("tomaarsen/span-marker-bert-base-acronyms")
# Run inference
entities = model.predict("Compression algorithms like Principal Component Analysis (PCA) can reduce noise and complexity.")
Downstream Use
You can finetune this model on your own dataset.
from span_marker import SpanMarkerModel, Trainer
# Download from the 🤗 Hub
model = SpanMarkerModel.from_pretrained("tomaarsen/span-marker-bert-base-acronyms")
# Specify a Dataset with "tokens" and "ner_tag" columns
dataset = load_dataset("conll2003") # For example CoNLL2003
# Initialize a Trainer using the pretrained model & dataset
trainer = Trainer(
model=model,
train_dataset=dataset["train"],
eval_dataset=dataset["validation"],
)
trainer.train()
trainer.save_model("tomaarsen/span-marker-bert-base-acronyms-finetuned")
Training Details
Training Set Metrics
| Training set | Min | Median | Max |
|---|---|---|---|
| Sentence length | 4 | 32.3372 | 170 |
| Entities per sentence | 0 | 2.6775 | 24 |
Training Hyperparameters
- learning_rate: 5e-05
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 2
Training Results
| Epoch | Step | Validation Loss | Validation Precision | Validation Recall | Validation F1 | Validation Accuracy |
|---|---|---|---|---|---|---|
| 0.3101 | 200 | 0.0083 | 0.9170 | 0.8894 | 0.9030 | 0.9766 |
| 0.6202 | 400 | 0.0063 | 0.9329 | 0.9149 | 0.9238 | 0.9807 |
| 0.9302 | 600 | 0.0060 | 0.9279 | 0.9338 | 0.9309 | 0.9819 |
| 1.2403 | 800 | 0.0058 | 0.9406 | 0.9092 | 0.9247 | 0.9812 |
| 1.5504 | 1000 | 0.0056 | 0.9453 | 0.9155 | 0.9302 | 0.9825 |
| 1.8605 | 1200 | 0.0054 | 0.9411 | 0.9271 | 0.9340 | 0.9831 |
Environmental Impact
Carbon emissions were measured using CodeCarbon.
- Carbon Emitted: 0.031 kg of CO2
- Hours Used: 0.204 hours
Training Hardware
- On Cloud: No
- GPU Model: 1 x NVIDIA GeForce RTX 3090
- CPU Model: 13th Gen Intel(R) Core(TM) i7-13700K
- RAM Size: 31.78 GB
Framework Versions
- Python: 3.9.16
- SpanMarker: 1.3.1.dev
- Transformers: 4.30.0
- PyTorch: 2.0.1+cu118
- Datasets: 2.14.0
- Tokenizers: 0.13.2
Citation
BibTeX
@software{Aarsen_SpanMarker,
author = {Aarsen, Tom},
license = {Apache-2.0},
title = {{SpanMarker for Named Entity Recognition}},
url = {https://github.com/tomaarsen/SpanMarkerNER}
}