metadata

library_name: transformers
license: apache-2.0
base_model: answerdotai/ModernBERT-base
tags:
  - generated_from_trainer
datasets:
  - conll2003
metrics:
  - precision
  - recall
  - f1
  - accuracy
model-index:
  - name: modernbert-ner-conll2003
    results:
      - task:
          name: Token Classification
          type: token-classification
        dataset:
          name: conll2003
          type: conll2003
          config: conll2003
          split: validation
          args: conll2003
        metrics:
          - name: Precision
            type: precision
            value: 0.8349195930423368
          - name: Recall
            type: recall
            value: 0.856277347694379
          - name: F1
            type: f1
            value: 0.8454636091724825
          - name: Accuracy
            type: accuracy
            value: 0.9751567306569059
language:
  - en
pipeline_tag: token-classification

ModernBERT NER (CoNLL2003)

This model is a fine-tuned version of answerdotai/ModernBERT-base on the conll2003 dataset for Named Entity Recognition (NER).

Robust performance on tasks involving the recognition of Persons, Organizations, and Locations.

It achieves the following results on the evaluation set:

Loss: 0.0992
Precision: 0.8349
Recall: 0.8563
F1: 0.8455
Accuracy: 0.9752

Model Details

Base Model: ModernBERT: https://doi.org/10.48550/arXiv.2412.13663.
Fine-tuning Dataset: CoNLL2003: https://huggingface.co./datasets/eriktks/conll2003.
Task: Named Entity Recognition (NER)

Training Data

The model is fine-tuned on the CoNLL2003 dataset, a well-known benchmark for NER. This dataset provides a solid foundation for the model to generalize on general English text.

Example Usage

Below is an example of how to use the model with the Hugging Face Transformers library:

from transformers import pipeline

ner = pipeline(task="token-classification", model="IsmaelMousa/modernbert-ner-conll2003", aggregation_strategy="max")

results = ner("Hi, I'm Ismael Mousa from Palestine working for NVIDIA inc.")

for entity in results:
    for key, value in entity.items():
        if key == "entity_group":
            print(f"{entity['word']} => {entity[key]}")

Results:

Ismael Mousa => PER
Palestine => LOC
NVIDIA => ORG

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 1e-06
train_batch_size: 8
eval_batch_size: 8
seed: 42
optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
0.2306	1.0	1756	0.2243	0.6074	0.6483	0.6272	0.9406
0.1415	2.0	3512	0.1583	0.7258	0.7536	0.7394	0.9583
0.1143	3.0	5268	0.1335	0.7731	0.7989	0.7858	0.9657
0.0913	4.0	7024	0.1145	0.7958	0.8256	0.8104	0.9699
0.0848	5.0	8780	0.1079	0.8120	0.8408	0.8261	0.9720
0.0728	6.0	10536	0.1036	0.8214	0.8452	0.8331	0.9730
0.0623	7.0	12292	0.1032	0.8258	0.8487	0.8371	0.9737
0.0599	8.0	14048	0.0990	0.8289	0.8527	0.8406	0.9745
0.0558	9.0	15804	0.0998	0.8331	0.8541	0.8434	0.9750
0.0559	10.0	17560	0.0992	0.8349	0.8563	0.8455	0.9752

Framework versions

Transformers 4.48.0.dev0
Pytorch 2.2.1+cu121
Datasets 3.2.0
Tokenizers 0.21.0