| --- |
| license: mit |
| datasets: |
| - ontonotes/conll2012_ontonotesv5 |
| language: |
| - en |
| base_model: |
| - google-bert/bert-large-cased |
| pipeline_tag: token-classification |
| --- |
| |
| # BERT-large-cased fine-tuned on OntoNotes 5.0 |
|
|
| This model is a fine-tuned version of [google-bert/bert-large-cased](https://huggingface.co/google-bert/bert-large-cased) on the English subset of the **OntoNotes 5.0** (CoNLL-2012) dataset. It is optimized for high-precision Named Entity Recognition (NER) across 18 entity categories. |
|
|
| ## 📊 Performance |
| The model achieves the following results on the OntoNotes 5.0 test set: |
|
|
| | **Entity** | **Precision** | **Recall** | **F1-Score** | **Support** | |
| | :--- | :---: | :---: | :---: | :---: | |
| | CARDINAL | 0.7813 | 0.7891 | 0.7851 | 1005 | |
| | DATE | 0.7988 | 0.8516 | 0.8244 | 1786 | |
| | EVENT | 0.5619 | 0.6941 | 0.6211 | 85 | |
| | FAC | 0.6880 | 0.5772 | 0.6277 | 149 | |
| | GPE | 0.9185 | 0.9207 | 0.9196 | 2546 | |
| | LANGUAGE | 0.8421 | 0.7273 | 0.7805 | 22 | |
| | LAW | 0.4762 | 0.6818 | 0.5607 | 44 | |
| | LOC | 0.6337 | 0.7163 | 0.6725 | 215 | |
| | MONEY | 0.8636 | 0.9099 | 0.8861 | 355 | |
| | NORP | 0.8481 | 0.8909 | 0.8690 | 990 | |
| | ORDINAL | 0.7054 | 0.7633 | 0.7332 | 207 | |
| | ORG | 0.8690 | 0.9046 | 0.8864 | 2002 | |
| | PERCENT | 0.8467 | 0.8821 | 0.8640 | 407 | |
| | PERSON | 0.9090 | 0.9217 | 0.9153 | 2134 | |
| | PRODUCT | 0.6667 | 0.6889 | 0.6776 | 90 | |
| | QUANTITY | 0.6972 | 0.6471 | 0.6712 | 153 | |
| | TIME | 0.6106 | 0.6133 | 0.6120 | 225 | |
| | WORK_OF_ART | 0.6354 | 0.6805 | 0.6571 | 169 | |
| | **micro avg** | **0.8412** | **0.8675** | **0.8542** | **12584** | |
| | **macro avg** | **0.7418** | **0.7700** | **0.7535** | **12584** | |
| | **weighted avg** | **0.8427** | **0.8675** | **0.8546** | **12584** | |
|
|
| ## 🛠 Training Details |
| - **Architecture**: `BertForTokenClassification` (Large) |
| - **Tokenizer**: `BertTokenizerFast` (using `is_split_into_words=True`) |
| - **Epochs**: 5 |
| - **Learning Rate**: 1e-5 |
| - **Batch Size**: 4 per device (2x V100 GPUs) |
| - **Gradient Accumulation**: 4 steps (Effective Batch Size = 32) |
| - **Max Sequence Length**: 128 |
| - **Weight Decay**: 0.01 |
| - **Mixed Precision (FP16)**: Enabled |
|
|
| ## 📂 Labels Mapping |
| The model identifies 18 entity types from OntoNotes 5.0: |
| `CARDINAL`, `DATE`, `EVENT`, `FAC`, `GPE`, `LANGUAGE`, `LAW`, `LOC`, `MONEY`, `NORP`, `ORDINAL`, `ORG`, `PERCENT`, `PERSON`, `PRODUCT`, `QUANTITY`, `TIME`, `WORK_OF_ART`. |
|
|
| ## 📂 Project Assets |
| - **GitHub Repository**: https://github.com/Learnrr/ontonotes5_ner_evaluation.git |
|
|
| | **Asset** | **File** | **Description** | |
| | :--- | :--- | :--- | |
| | **Model Weights** | `model.safetensors` | Large-scale checkpoint (~1.2 GB). | |
| | **Configuration** | `config.json` | Model architecture & `id2label` mapping. | |
| | **Vocabulary** | `vocab.txt` | BERT-cased specific vocabulary. | |
| | **Tokenizer** | `tokenizer.json` | Optimized fast tokenizer configuration. | |
| | **Special Tokens** | `special_tokens_map.json` | Definitions for BOS, EOS, and Padding tokens. | |
| | **Training Args** | `training_args.bin` | Detailed hyperparameter dump from the Trainer. | |
|
|
| ## 🚀 Usage |
| ```python |
| from transformers import pipeline |
| |
| model_checkpoint = "learnrr/bert-large-ontonotes5-ner" |
| token_classifier = pipeline( |
| "token-classification", |
| model=model_checkpoint, |
| aggregation_strategy="simple" |
| ) |
| |
| text = "The United Nations is headquartered in New York City." |
| results = token_classifier(text) |
| |
| for entity in results: |
| print(f"Entity: {entity['word']} | Label: {entity['entity_group']} | Score: {entity['score']:.4f}") |
| ``` |
