Model card: throughput across CPU, MPS and CUDA

Complete CPU run: all 16 models on all three devices, and the throughput finding

Model card: Evaluation and Device Comparison section

Add CPU to the device comparison and a generated evaluation figure across CUDA, CPU and MPS

Sync repository at da4bb2e

Sync docs/

Model card: what the artifacts are, what the RTX 5090 vs M1 Ultra comparison found, and how to reproduce it

Remove viewer/node_modules: a nested path my exclude pattern missed

CUDA benchmarks on an RTX 5090: checkpoints, agreement runs, shared init weights, and the corrected comparison

Upload folder using huggingface_hub

Upload folder using huggingface_hub

initial commit