madhavgohel/bert-token-onnx

🔍 BERT Token Classification – Important Chunk Extractor (ONNX) - InTextIQ

This model identifies and extracts important parts of input sentences using BERT-based token classification, exported to the ONNX format for optimized inference.

🧠 Use Case

This model is designed for context engineering — to extract semantically important words or chunks from sentences or chat messages, enabling better personalization in downstream applications like AI assistants or dialogue systems.

Example:

Input:  I’ll be unavailable tomorrow due to a team offsite.

Output: [unavailable, tomorrow, team offsite]

---

🛠️ Model Details

• Architecture: BERT (bert-base-uncased) fine-tuned for token classification

• Exported to: ONNX for efficient runtime inference via Optimum

• Labels:

  label_list = ["O", "B-IMPORTANT", "I-IMPORTANT"]

---

📦 How to Use (with 🤗 Transformers + Optimum)

from transformers import AutoTokenizer
from optimum.onnxruntime import ORTModelForTokenClassification
import torch

model = ORTModelForTokenClassification.from_pretrained("madhavgohel/bert-token-onnx", file_name="model.onnx")
tokenizer = AutoTokenizer.from_pretrained("madhavgohel/bert-token-onnx")

text = "I'm a software engineer with 5 years experience looking to switch to a data science role."

inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs)
predictions = torch.argmax(outputs.logits, dim=-1)

tokens = tokenizer.convert_ids_to_tokens(inputs["input_ids"][0])
important_tokens = [tok for tok, label in zip(tokens, predictions[0]) if label == 1]
print("Important tokens:", important_tokens)

---

📁 Files Included

File	Purpose
model.onnx	Exported ONNX model
config.json	Model config
tokenizer_config.json	Tokenizer config
vocab.txt	Vocabulary for BERT tokenizer
special_tokens_map.json	Tokenization map for special tokens
README.md	Model usage documentation