finetuned-distilbert-connllp

Model Card for Fine-tuned DistilBERT on CoNLL-PP for Named Entity Recognition

Model Details

Model Description

This model is a fine-tuned version of DistilBERT-Base-Cased for Named Entity Recognition (NER) using the CoNLL-PP dataset. The model identifies entities in text such as locations, organizations, persons, etc.

• Developed by: shogun-the-great
• Model type: Token Classification (NER)
• Language(s): English
• License: Apache-2.0 (or specify your license)
• Finetuned from model: distilbert-base-cased

Model Sources

• Dataset: conllpp

Uses

Direct Use

This model can be directly used for Named Entity Recognition tasks to extract entities such as locations, organizations, and persons from text. Typical use cases include:

• NER for document analysis.
• Text classification and entity extraction in information retrieval systems.
• Integration with chatbots to identify user-specific entities.

Downstream Use

This model can be further fine-tuned for specific tasks requiring NER in specific domains like medical, legal, etc.

Out-of-Scope Use

This model might not perform well for:

• Non-English text.
• Extremely noisy or unstructured text where entities are not clearly defined.

Bias, Risks, and Limitations

Bias

The model’s predictions are influenced by the dataset used during fine-tuning. If the dataset contains biases, they may be reflected in the predictions.

Risks

• False positives: Incorrectly identified entities (e.g., non-person names identified as persons).
• False negatives: Important entities missed.
• Limited generalization to non-CoNLL-PP entity types or domains.

Recommendations

• Regularly update the model with new data for better generalization.
• Review and monitor predictions to ensure the accuracy of identified entities.

How to Get Started with the Model

You can load the fine-tuned model directly from the Hugging Face Hub:

from transformers import AutoTokenizer, AutoModelForTokenClassification

# Load the tokenizer and model from Hugging Face Hub
model_name = "shogun-the-great/finetuned-distilbert-connllp"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)

# Example usage for NER
text = "Barack Obama was born in Hawaii."

inputs = tokenizer(text, return_tensors="pt", truncation=True, is_split_into_words=True)
outputs = model(**inputs)

# Get the predicted labels for each token
predictions = outputs.logits.argmax(dim=-1)
tokens = inputs.tokens()

# Convert predictions to entity names
predicted_tags = [model.config.id2label[prediction.item()] for prediction in predictions[0]]
for token, tag in zip(tokens, predicted_tags):
    print(f"{token}: {tag}")