README.md

metadata

license: apache-2.0
datasets:
  - custom
  - chatgpt
language:
  - en
metrics:
  - precision
  - recall
  - f1
  - accuracy
pipeline_tag: token-classification
library_name: transformers
new_version: v1.1
tags:
  - token-classification
  - ner
  - named-entity-recognition
  - text-classification
  - sequence-labeling
  - transformer
  - bert
  - nlp
  - pretrained-model
  - dataset-finetuning
  - deep-learning
  - huggingface
  - conll2012
  - real-time-inference
  - efficient-nlp
  - high-accuracy
  - gpu-optimized
  - chatbot
  - information-extraction
  - search-enhancement
  - knowledge-graph
  - legal-nlp
  - medical-nlp
  - financial-nlp
base_model:
  - boltuix/bert-lite

🌟 Boltuix BERT-NER Model 🌟

🚀 Model Details

🌈 Description

• ✨ Fine-tuned for Named Entity Recognition (NER)
• 📚 Dataset: CoNLL-2012
• 🔍 Recognizes 37 entity types across diverse domains like people, places, organizations, laws, events, and more!
• 💬 Works great for sentence-level and document-level tagging in English.
• 🧠 Training examples: 115,812 | ✅ Validation: 15,680 | 🧪 Test: 12,217

🔧 Info

• Developer: Boltuix 🧙‍♂️
• Fuel: Passion 🧠
• License: Apache 2.0 📜
• Language: English 🇬🇧
• Type: Transformer-based Token Classification 🤖
• Version: v1.0 🎈
• Trained: Before March 27, 2025

🔗 Links

• 🧠 Model Repo
• 📄 CoNLL-2012 Paper
• 💡 Demo: Coming Soon

---

🎯 Use Cases for NER

🌟 Direct Applications

• Extracting names, places, and dates from news, blogs, and reports
• Powering chatbots with contextual awareness
• Enhancing search with semantic understanding
• Building dynamic knowledge graphs

🌱 Downstream Tasks

• Medical & legal domain adaptation
• Multilingual extensions (with retraining)
• Custom entity sets for finance, e-commerce, etc.

❌ Limitations

• 📌 English-only out of the box
• 🚫 May not generalize to informal, low-resource, or code-mixed texts
• ⚖️ May reflect dataset bias (CoNLL-2012 is newswire-heavy)

---

🛠️ Getting Started

🧪 Inference Code

from transformers import AutoTokenizer, AutoModelForTokenClassification
import torch

tokenizer = AutoTokenizer.from_pretrained("boltuix/bert-ner")
model = AutoModelForTokenClassification.from_pretrained("boltuix/bert-ner")

text = "Barack Obama visited Microsoft headquarters in Seattle."
inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs)
predictions = outputs.logits.argmax(dim=-1)

tokens = tokenizer.convert_ids_to_tokens(inputs["input_ids"][0])
label_map = model.config.id2label
labels = [label_map[p.item()] for p in predictions[0]]

for token, label in zip(tokens, labels):
    if token not in tokenizer.all_special_tokens:
        print(f"{token:15} → {label}")

✨ Example Output

barack          → B-PERSON
obama           → I-PERSON
visited         → O
microsoft       → B-ORG
headquarters    → O
in              → O
seattle         → B-GPE
.               → O

---

🧠 Entity Labels (CoNLL-2012)

Here are all 37 labels supported by the model:

🔹 O – Outside

🔢 Beginning (B-) Tags

🔢 B-CARDINAL – CARDINAL
📅 B-DATE – DATE
🎉 B-EVENT – EVENT
🏢 B-FAC – FACILITY
🌍 B-GPE – COUNTRY/CITY
🗣️ B-LANGUAGE – LANGUAGE
⚖️ B-LAW – LAW
🗺️ B-LOC – LOCATION
💰 B-MONEY – MONEY
🧑‍🤝‍🧑 B-NORP – GROUP
🔟 B-ORDINAL – ORDINAL
🏛️ B-ORG – ORGANIZATION
📊 B-PERCENT – PERCENT
👤 B-PERSON – PERSON
📦 B-PRODUCT – PRODUCT
📏 B-QUANTITY – QUANTITY
⏰ B-TIME – TIME
🎨 B-WORK_OF_ART – WORK_OF_ART

🔢 Inside (I-) Tags

🔢 I-CARDINAL
📅 I-DATE
🎉 I-EVENT
🏢 I-FAC
🌍 I-GPE
🗣️ I-LANGUAGE
⚖️ I-LAW
🗺️ I-LOC
💰 I-MONEY
🧑‍🤝‍🧑 I-NORP
🔟 I-ORDINAL
🏛️ I-ORG
📊 I-PERCENT
👤 I-PERSON
📦 I-PRODUCT
📏 I-QUANTITY
⏰ I-TIME
🎨 I-WORK_OF_ART

---

📈 Performance

Metric	Score
🎯 Precision	0.85
🕸️ Recall	0.87
🎶 F1 Score	0.86
✅ Accuracy	0.92

• 📊 Evaluation tool: seqeval
• 🧪 Dataset: CoNLL-2012 test split

---

⚙️ Training Setup

• 💻 Hardware: NVIDIA GPU
• ⏱️ Training Time: ~2 hours
• 🐘 Parameters: ~11M
• 🎛️ Optimizer: AdamW (default)
• 📦 Mixed precision: No (fp32)

---

🌍 Carbon Impact

• 💻 Trained Locally
• ☁️ Region: Boltuix’s Base
• 🌱 Emissions: ~50g CO₂eq
• 📊 Measured via: ML Impact

---

✍️ Contact

• 🧠 Author: Boltuix
• 📬 Email: [email protected]