Benchmark Comparison

Performance comparisons between TinyLoop and generic runtimes for weight-shared looped transformer inference.

• Measured Benchmarks — directly measured TinyLoop numbers on H100 and RTX 4090
• vs Generic Runtimes — why vLLM, TensorRT-LLM, and llama.cpp lose on looped checkpoints
• Claims & Data Sources — what's safe to claim, what's not, and where the data lives