Hugging Face's logo

Aaron-96
/
multiNERD_fine-tuned_only_English_roberta

Token Classification
SpanMarker
PyTorch
ner
named-entity-recognition
generated_from_span_marker_trainer
Eval Results
Model card Files Community
multiNERD_fine-tuned_only_English_roberta / README.md
Aaron-96's picture
Aaron-96
Upload 10 files
c298231
preview code
|
raw
history blame contribute delete
8.69 kB
metadata
library_name: span-marker
tags:
  - span-marker
  - token-classification
  - ner
  - named-entity-recognition
  - generated_from_span_marker_trainer
metrics:
  - precision
  - recall
  - f1
widget:
  - text: >-
      The Bengal tiger is the most common subspecies of tiger, constituting
      approximately 80% of the entire tiger population, and is found in
      Bangladesh, Bhutan, Myanmar, Nepal, and India.
  - text: >-
      In other countries, it is a non-commissioned rank (e.g. Spain, Italy,
      France, the Netherlands and the Indonesian Police ranks).
  - text: >-
      The filling consists of fish, pork and bacon, and is seasoned with salt
      (unless the pork is already salted).
  - text: >-
      This stood until August 20, 1993 when it was beaten by one 1 / 100th of a
      second by Colin Jackson of Great Britain in Stuttgart, Germany, a
      subsequent record that stood for 13 years.
  - text: >-
      Ann Patchett ’s novel " Bel Canto ", was another creative influence that
      helped her manage a plentiful cast of characters.
pipeline_tag: token-classification
model-index:
  - name: SpanMarker
    results:
      - task:
          type: token-classification
          name: Named Entity Recognition
        dataset:
          name: Unknown
          type: unknown
          split: eval
        metrics:
          - type: f1
            value: 0.9130661114003124
            name: F1
          - type: precision
            value: 0.9148758606300855
            name: Precision
          - type: recall
            value: 0.9112635078969243
            name: Recall

SpanMarker

This is a SpanMarker model that can be used for Named Entity Recognition.

Model Details

Model Description

  • Model Type: SpanMarker
  • Maximum Sequence Length: 256 tokens
  • Maximum Entity Length: 6 words

Model Sources

Model Labels

Label Examples
ANIM "vertebrate", "moth", "G. firmus"
BIO "Aspergillus", "Cladophora", "Zythiostroma"
CEL "pulsar", "celestial bodies", "neutron star"
DIS "social anxiety disorder", "insulin resistance", "Asperger syndrome"
EVE "Spanish Civil War", "National Junior Angus Show", "French Revolution"
FOOD "Neera", "Bellini ( cocktail )", "soju"
INST "Apple II", "Encyclopaedia of Chess Openings", "Android"
LOC "Kīlauea", "Hungary", "Vienna"
MEDIA "CSI : Crime Scene Investigation", "Big Comic Spirits", "American Idol"
MYTH "Priam", "Oźwiena", "Odysseus"
ORG "San Francisco Giants", "Arm Holdings", "RTÉ One"
PER "Amelia Bence", "Tito Lusiardo", "James Cameron"
PLANT "vernal squill", "Sarracenia purpurea", "Drosera rotundifolia"
TIME "prehistory", "Age of Enlightenment", "annual paid holiday"
VEHI "Short 360", "Ferrari 355 Challenge", "Solution F / Chretien Helicopter"

Uses

Direct Use for Inference 

from span_marker import SpanMarkerModel

# Download from the 🤗 Hub
model = SpanMarkerModel.from_pretrained("span_marker_model_id")
# Run inference
entities = model.predict("Ann Patchett ’s novel \" Bel Canto \", was another creative influence that helped her manage a plentiful cast of characters.")

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("span_marker_model_id")

# 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("span_marker_model_id-finetuned")

Training Details

Training Set Metrics

Training set Min Median Max
Sentence length 2 21.6493 237
Entities per sentence 0 1.5369 36

Training Hyperparameters

  • learning_rate: 1e-05
  • train_batch_size: 16
  • eval_batch_size: 16
  • 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: 1
  • mixed_precision_training: Native AMP

Training Results

Epoch Step Validation Loss Validation Precision Validation Recall Validation F1 Validation Accuracy
0.0576 1000 0.0142 0.8714 0.7729 0.8192 0.9698
0.1153 2000 0.0107 0.8316 0.8815 0.8558 0.9744
0.1729 3000 0.0092 0.8717 0.8797 0.8757 0.9780
0.2306 4000 0.0082 0.8811 0.8886 0.8848 0.9798
0.2882 5000 0.0084 0.8523 0.9163 0.8831 0.9790
0.3459 6000 0.0079 0.8700 0.9113 0.8902 0.9802
0.4035 7000 0.0070 0.9107 0.8859 0.8981 0.9822
0.4611 8000 0.0069 0.9259 0.8797 0.9022 0.9827
0.5188 9000 0.0067 0.9061 0.8965 0.9013 0.9829
0.5764 10000 0.0066 0.9034 0.8996 0.9015 0.9829
0.6341 11000 0.0064 0.9160 0.8996 0.9077 0.9839
0.6917 12000 0.0066 0.8952 0.9121 0.9036 0.9832
0.7494 13000 0.0062 0.9165 0.9009 0.9086 0.9841
0.8070 14000 0.0062 0.9010 0.9121 0.9065 0.9835
0.8647 15000 0.0062 0.9084 0.9127 0.9105 0.9842
0.9223 16000 0.0060 0.9151 0.9098 0.9125 0.9846
0.9799 17000 0.0060 0.9149 0.9113 0.9131 0.9848

Framework Versions

  • Python: 3.8.16
  • SpanMarker: 1.5.0
  • Transformers: 4.29.0.dev0
  • PyTorch: 1.10.1
  • Datasets: 2.15.0
  • 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}
}