lila-e8-time2026

Lila-E8 trained on Time-2026 temporal lattice stories

Model Details

• Architecture: Lila-E8 (Lie Lattice Attention)
• E8 Quantization: 240 roots
• Geometric Attention: E8-biased
• Training Steps: 9000
• Framework: PyTorch

Usage

import torch
from model.lila_e8 import LilaE8Config, LilaE8

# Load checkpoint
checkpoint = torch.load('checkpoint_step_9000.pt', map_location='cpu')
config = checkpoint['config']
model = LilaE8(config)
model.load_state_dict(checkpoint['model_state_dict'])

# Generate
# (see inference examples in repo)

Training

Trained on NVIDIA RTX 3080 Ti using Nix reproducible environment.

See sovereign-lila-e8 for training code.