KevinDavidHayes/regressor-entropy

Long-Context Attention Regressor (Entropy)

Predicts the attention entropy of a text sample - how spread out vs focused the attention patterns are.

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model = AutoModelForSequenceClassification.from_pretrained("KevinDavidHayes/regressor-entropy")
tokenizer = AutoTokenizer.from_pretrained("KevinDavidHayes/regressor-entropy")

text = "Your text here..."
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=8192)

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

# Higher score = more spread attention (uses more context)

Training

• Base model: ModernBERT-base (8K context)
• Target: Normalized attention entropy
• Labels: Generated using Qwen2.5-7B-Instruct attention analysis at layer 14

Citation

Part of research on attention-based data filtering for long-context pretraining.