Update benchmarks for v2 (joint cross-modal training)

Update model: v2 with joint cross-modal training (+40% I↔A)

Remove throughput and efficiency tables

Add MSCOCO Karpathy 5K benchmark results for all models

Add ImageBind and EBind throughput to all tables

Add efficiency comparison table, improve throughput visibility

Add LCO-3B and Nemotron-3B benchmark comparisons + throughput table

Add augmem.ai and GitHub links

Add model weights

Add model card

initial commit