AbstractPhil / geo-beatrix-resnet

geo-beatrix-resnet34-step12-feats100

Geometric Basin Classification for CIFAR-100

🚧 Training in Progress 🚧

Current Status: Epoch 150/200

Current Performance

All Training Runs

Comparison to State-of-the-Art

🎯 Current target : Beat geo-beatrix (69.0%) - Currently -12.88%

Architecture

• Base : ResNet34 (torchvision)
• Pretrained : From scratch
• Features : 512-dim from ResNet34
• Positional Encoding : Devil's Staircase (Cantor function, 1883)
• PE Levels : 12
• PE Features/Level : 100
• Classification : Geometric Basin Compatibility (NO cross-entropy)
• Attention Mechanisms : NONE
• Mixing : Fractal (triadic multi-patch)

Training Configuration

{
  "model_name": "geo-beatrix-resnet34-step12-feats100",
  "model_type": "geometric_basin_classifier",
  "num_classes": 100,
  "batch_size": 1024,
  "num_epochs": 200,
  "base_learning_rate": 0.001,
  "weight_decay": 0.0,
  "warmup_epochs": 10,
  "pe_levels": 12,
  "pe_features_per_level": 100,
  "dropout": 0.1,
  "pretrained_resnet": false,
  "frozen_resnet": false,
  "a100_optimizations": {
    "mixed_precision": true,
    "torch_compile": false,
    "channels_last": true,
    "gradient_checkpointing": false
  },
  "alphamix": {
    "enabled": true,
    "fractal_mode": true,
    "range": [
      0.3,
      0.7
    ],
    "spatial_ratio": 0.05,
    "curriculum_start": 0.25,
    "curriculum_end": 0.5,
    "fractal_steps": [
      1,
      3
    ],
    "fractal_scales": [
      0.3333333333333333,
      0.1111111111111111,
      0.037037037037037035
    ]
  },
  "architecture": "ResNet34 + Devil's Staircase PE",
  "loss_function": "Geometric Basin Compatibility",
  "cross_entropy": false,
  "attention_mechanisms": false,
  "timestamp": "20251010_234239"
}

Files Structure

├── model.pt                 (BEST overall model - easy access!)
├── model.safetensors        (BEST overall model - easy access!)
├── best_model_info.json     (which epoch/run this came from)
├── runs_history.json        (all training runs and their results)
├── README.md
├── weights/geo-beatrix-resnet34-step12-feats100/20251010_234239/
│   ├── model.pt                 (best from this training run)
│   ├── model.safetensors        (best from this training run)
│   ├── config.json
│   ├── training_log.txt
│   └── checkpoints/
│       ├── checkpoint_epoch_50.safetensors
│       ├── checkpoint_epoch_100.safetensors
│       └── checkpoint_epoch_150.safetensors
│       (snapshots every 10 epochs)
└── runs/geo-beatrix-resnet34-step12-feats100/20251010_234239/
    ├── events.out.tfevents.*    (TensorBoard logs)
    └── metrics.csv              (training metrics)

Note : The root model.pt and model.safetensors always contain the best model across all training runs!

Usage

from huggingface_hub import hf_hub_download
import torch

# EASIEST: Download BEST overall model from root (recommended!)
from safetensors.torch import load_file
model_path = hf_hub_download(
    repo_id="AbstractPhil/geo-beatrix-resnet",
    filename="model.safetensors"
)
state_dict = load_file(model_path)
# model.load_state_dict(state_dict)

# Check which epoch/run the best model came from
info_path = hf_hub_download(
    repo_id="AbstractPhil/geo-beatrix-resnet",
    filename="best_model_info.json"
)
with open(info_path) as f:
    best_info = json.load(f)
    print(f"Best model: epoch {best_info['epoch']}, {best_info['test_accuracy']:.2f}%")

# Or download from specific training run
model_path = hf_hub_download(
    repo_id="AbstractPhil/geo-beatrix-resnet",
    filename="weights/geo-beatrix-resnet34-step12-feats100/20251010_234239/model.safetensors"
)

# Download specific epoch checkpoint
epoch_checkpoint = hf_hub_download(
    repo_id="AbstractPhil/geo-beatrix-resnet",
    filename="weights/geo-beatrix-resnet34-step12-feats100/20251010_234239/checkpoints/checkpoint_epoch_100.safetensors"
)

Training History

Best Checkpoint

• Epoch: 160
• Train Acc: 74.61%
• Test Acc: 56.12%
• Alpha: 0.4855
• Loss: 0.0000

Latest 5 Epochs

• Epoch 146 : Train 71.93%, Test 54.14%, α=0.4849, Loss=0.7198
• Epoch 147 : Train 73.77%, Test 0.00%, α=0.4851, Loss=0.7246
• Epoch 148 : Train 72.73%, Test 53.57%, α=0.4852, Loss=0.7322
• Epoch 149 : Train 70.84%, Test 0.00%, α=0.4854, Loss=0.6536
• Epoch 150 : Train 74.61%, Test 53.12%, α=0.4855, Loss=0.7047

Training Milestones

• 🎯 50% Accuracy reached at epoch 66
• 📊 α ≥ 0.40 reached at epoch 36
• 📊 α ≥ 0.44 (near triadic equilibrium) at epoch 59

Innovation

✅ NO attention mechanisms
✅ NO cross-entropy loss
✅ Fractal positional encoding (Cantor function from 1883)
✅ Geometric compatibility classification
✅ ResNet34 backbone (proven CNN architecture) ✅ Triadic fractal mixing (base-3 aligned)

Repository : https://huggingface.co/AbstractPhil/geo-beatrix-resnet
Author : AbstractPhil
Framework : PyTorch