v3: patch_size=4 (64 tokens), 2 core layers, iters [2,3,4], ~16min total training

PERF: force fp32 for FFT, JIT scan, fix RMSNorm, cache Manhattan

Fix: force_download=True to bypass HF cache, add assertion checks

v2: Use pre-trained SD-VAE, fix all bugs, pre-cache everything, massive speedup

v2: Use pre-trained SD-VAE, fix all bugs, pre-cache everything, massive speedup

Perf fix: cache Manhattan dist, optimize scan, pre-cache CLIP, fix deprecated AMP API

Perf fix: cache Manhattan dist, optimize scan, pre-cache CLIP, fix deprecated AMP API

Update README with Colab notebook link

Add complete Colab/Kaggle training notebook

Add ARCHITECTURE.md

Add test_iris.py

Add train_iris.py

Add iris_model.py

Add IRIS architecture - novel mobile-first image generation

initial commit