agentlans / bge-small-en-region-classifier

bge-small-en-region-classifier

This model classifies English text into its most likely geographic region based on linguistic nuances, context, and terminology.

Supported Regions

Intended Uses & Limitations

Intended Uses

• Content Localization: Route user feedback, support tickets, or inquiries to the appropriate regional teams automatically.
• Content Moderation & Compliance: Filter or flag regional text variations for compliance with local digital standards or cultural sensitivities.
• Market Research & Analytics: Analyze large-scale social media data, product reviews, or forum posts to understand the geographic distribution of your audience.
• Dataset Auditing: Profile large text datasets to ensure geographic diversity or to detect regional imbalances before training downstream models.

Limitations & Biases

• Linguistic vs. Topical Clues: The model relies on regional slang, sentence structure, specific spellings (e.g., colour vs. color ), and language pulled from localized websites. Text written in highly standardized "Global English" will likely default to Other . Additionally, while named entities were anonymized in the training data, strong topical associations (e.g., discussions about cricket, kangaroos, or Thanksgiving) can still skew predictions toward a specific region.
• Code-Switching: Performance may be impacted by text that heavily mixes English with local languages (e.g., Hinglish, Singlish). While these mixed languages are strong indicators of a broader geographic area (like South or Southeast Asia), heavy code-switching can lower specific classification accuracy.
• Short Inputs: Very short texts (e.g., single-word queries or brief phrases) often lack enough linguistic signals for accurate classification, unless they contain a highly specific regional term.
• Demographic & Dialect Bias: The model predicts based on dominant regional linguistic patterns and may misclassify text written by regional minorities, expatriates, or non-native speakers. Furthermore, several major English variations and dialects—such as Caribbean English, African-American Vernacular English (AAVE), and various English-based pidgins—were excluded from the training data.
• Not a Geolocation Tool: This model predicts the linguistic style associated with a broad region, not a user's physical GPS location. For example, a user physically located in Canada writing in British English will likely be classified under UK and Ireland .

Use

To get started with this model in Python using the Hugging Face Transformers library, run the following code:

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_id = "agentlans/bge-small-en-region-classifier"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)

text = "Replace this with your input text."
inputs = tokenizer(text, return_tensors="pt")

with torch.no_grad():
    logits = model(**inputs).logits

predicted_class_id = logits.argmax().item()
predicted_class_name = model.config.id2label[predicted_class_id]

print(f"Predicted Class ID: {predicted_class_id}")
print(f"Predicted Class Name: {predicted_class_name}")

Intended Uses & Limitations

Intended Use

This model is designed for sequence classification tasks. Below are the specific class labels mapped to their corresponding IDs:

Training Details

Hyperparameters

The following hyperparameters were used during fine-tuning:

• Learning Rate: 5e-05
• Train Batch Size: 8
• Eval Batch Size: 8
• Optimizer: OptimizerNames.ADAMW_TORCH_FUSED
• Number of Epochs: 3.0
• Mixed Precision: BF16

Training Results & Evaluation

During fine-tuning, the model achieved the following results on the evaluation set:

Test Dataset Performance

Classification Metrics

Confusion Matrix

Sample Predictions

Speed Performance

• Training Runtime: 5003.7568 seconds
• Train Samples per Second: 485.635
• Evaluation Runtime: 24.8749 seconds
• Eval Samples per Second: 1809.051

Framework Versions

• Transformers: 5.0.0.dev0
• PyTorch: 2.9.1+cu128