nbroad/span-marker-xdistil-l12-h384-orgs-v3

SpanMarker with microsoft/xtremedistil-l12-h384-uncased on FewNERD, CoNLL2003, and OntoNotes v5

This is a SpanMarker model trained on the FewNERD, CoNLL2003, and OntoNotes v5 dataset that can be used for Named Entity Recognition. This SpanMarker model uses microsoft/xtremedistil-l12-h384-uncased as the underlying encoder.

Model Details

Model Description

• Model Type: SpanMarker
• Encoder: microsoft/xtremedistil-l12-h384-uncased
• Maximum Sequence Length: 256 tokens
• Maximum Entity Length: 8 words
• Training Dataset: FewNERD, CoNLL2003, and OntoNotes v5
• Language: en
• License: cc-by-sa-4.0

Model Sources

• Repository: SpanMarker on GitHub
• Thesis: SpanMarker For Named Entity Recognition

Model Labels

Label	Examples
ORG	"Texas Chicken", "IAEA", "Church 's Chicken"

Evaluation

Metrics

Label	Precision	Recall	F1
all	0.7620	0.7498	0.7559
ORG	0.7620	0.7498	0.7559

Uses

Direct Use for Inference

from span_marker import SpanMarkerModel

# Download from the 🤗 Hub
model = SpanMarkerModel.from_pretrained("nbroad/span-marker-xdistil-l12-h384-orgs-v3")
# Run inference
entities = model.predict("SCL claims that its methodology has been approved or endorsed by agencies of the Government of the United Kingdom and the Federal government of the United States, among others.")

Downstream Use

You can finetune this model on your own dataset.

Click to expand

from span_marker import SpanMarkerModel, Trainer

# Download from the 🤗 Hub
model = SpanMarkerModel.from_pretrained("nbroad/span-marker-xdistil-l12-h384-orgs-v3")

# 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("nbroad/span-marker-xdistil-l12-h384-orgs-v3-finetuned")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Sentence length	1	23.5706	263
Entities per sentence	0	0.7865	39

Training Hyperparameters

• learning_rate: 0.0003
• train_batch_size: 128
• eval_batch_size: 128
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 3
• mixed_precision_training: Native AMP

Training Results

Epoch	Step	Validation Loss	Validation Precision	Validation Recall	Validation F1	Validation Accuracy
0.5720	600	0.0086	0.7150	0.7095	0.7122	0.9660
1.1439	1200	0.0074	0.7556	0.7253	0.7401	0.9682
1.7159	1800	0.0073	0.7482	0.7619	0.7550	0.9702
2.2879	2400	0.0072	0.7761	0.7573	0.7666	0.9713
2.8599	3000	0.0070	0.7691	0.7688	0.7689	0.9720

Framework Versions

• Python: 3.10.12
• SpanMarker: 1.5.0
• Transformers: 4.35.2
• PyTorch: 2.1.0a0+32f93b1
• Datasets: 2.15.0
• Tokenizers: 0.15.0

Citation

BibTeX

@software{Aarsen_SpanMarker,
    author = {Aarsen, Tom},
    license = {Apache-2.0},
    title = {{SpanMarker for Named Entity Recognition}},
    url = {https://github.com/tomaarsen/SpanMarkerNER}
}