Novel GPU memory reduction experiments

CogMemBench paper for Kaggle + GPU memory reduction experiments

GPU memory experiment + CogMemBench on real model

Update paper + README with final GPU results, fix Colab, research GPU memory reduction

Fix engine compilation errors

Test engine compilation

FigQuant GPU training test (with dtype fix)

Fix lowram backward dtype bug + AGPL licenses + rerun GPU test

Fix FigQuant GPU benchmark (use figcache mode) + test engine conversion

Engine Phase 3: Complete format converter + BPE tokenizer + kernel wiring

Engine Phase 2: Full transformer forward pass + tokenizer + attention

Engine Phase 1b: AVX2 vectorized matmul + RMSNorm kernels

Lila Engine Phase 1: Foundation — matmul kernel + model loader + token generation

Full GPU benchmark including FP32 quant quality test

Training-only GPU benchmark (skip FP32 quant quality — already proven on CPU)

Lila inference engine build plan + fix LilaCore

Lila restructure script

GPU benchmark script (fixed for PyTorch 2.11)

Update ML Intern artifact metadata

initial commit