metadata
tags:
- vision
- image-classification
- fractal-positional-encoding
- geometric-deep-learning
- devil-staircase
- simplex-geometry
license: mit
ViT-Beatrix: Fractal PE + Geometric Simplex Vision Transformer
This model integrates Devil's Staircase positional encoding with geometric simplex features for vision tasks. Trained on CIFAR-10.
Model Details
- Architecture: Vision Transformer with fractal positional encoding
- Dataset: CIFAR-100 (100 classes)
- Embedding Dimension: 512
- Depth: 4 layers
- Patch Size: 4x4
- PE Levels: 12
- Simplex Dimension: 5-simplex
Training
- Dataset: CIFAR-100
- Epochs: 2
- Best Accuracy: 0.1820
- Batch Size: 512
- Learning Rate: 0.001
Loss Components
- Task Loss Weight: 1.0
- Flow Alignment Weight: 0.5
- Coherence Weight: 0.3
- Multi-Scale Weight: 0.2
Usage
from geovocab2.train.model.vit_beatrix import SimplifiedGeometricClassifier
from safetensors.torch import load_file
# Load model
model = SimplifiedGeometricClassifier(
num_classes=100, # CIFAR-100
img_size=32,
embed_dim=512,
depth=4
)
# Load weights (renamed from model_best.safetensors to model.safetensors in Hub)
state_dict = load_file("model.safetensors")
model.load_state_dict(state_dict)
model.eval()
# Inference
output = model(images)
Citation
@misc{vit-beatrix,
author = {AbstractPhil},
title = {ViT-Beatrix: Fractal Positional Encoding with Geometric Simplices},
year = {2025},
url = {https://github.com/AbstractEyes/lattice_vocabulary}
}
License
MIT License