Card: base_model_relation: quantized (list under the base model's Quantizations)

Add measured Galaxy S26 NPU/GPU section

Card update: rank-4 re-export GPU results (full mobile delegation; measured on-device)

Rank-4 repeat_kv re-export: full mobile-GPU delegation (bitwise-identical CPU outputs)

Rank-4 repeat_kv re-export: full mobile-GPU delegation (bitwise-identical CPU outputs)

Card: add the measured Pixel 8a rows (GPU vs CPU) that the Android sweep produced

Card: measured Performance table (M4 Max + iPhone 17 Pro), verified runnable usage script, accuracy note

LFM2.5-Encoder-350M-Prompt-Router LiteRT: int8 (iPhone-verified bit-exact) + fp16, task-level parity verified

initial commit