Kernel work: 11L GQA U-NET BODY (x2): 11-layer dim-512 GQA U-Net transformer base with XSA, BigramHash, value embedding, and SmearGate. Specs: num_layers=11; model_dim=512; vocab_size=1024; num_heads=8; num_kv_heads=4.
Neural work: 10K-VOCAB CASEOPS LOOP BODY (x4): SP10240 CaseOps neural body: copy as an 11-layer, dim-512, looped GQA transformer with phased LoRA TTT packaging. Specs: num_layers=11; model_dim=512; vocab_size=10240; num_heads=8; num_kv_heads=4. LEADERBOARD TRANSFORMER VARIANTS (x3): architecture family detected in measured rows; exact flags are preserved in the row details. 11L GQA U-NET BODY (x3): 11-layer dim-512 GQA U-Net transformer base with XSA, BigramHash, value embedding, and SmearGate. Specs: num_layers=11; model_dim=512; vocab_size=1024; num_heads=8; num_kv_heads=4.
Week 07 covers 3 daily reports from 2026-04-29 through 2026-05-01. Measured surfaces: 3 daily datapoints across VAL_BPB (×2) · CUSTOM/FA3 FB RATIO (×1).
Techniques to copy/reference this week: SmearGate; SP10240 lossless-caps CaseOps tokenizer; pergroup compression; embed int7; matrix int6; phased LoRA TTT; 16 MB artifact cap.