Эксперимент: Balanced design attention (#444)

{ "worked": true, "confidence": 9, "verdict": "Built a readable MVP for exact cyclic block-design attention, including incidence construction, gather/local-attention/scatter averaging, and dense/random/contiguous comparisons. The core mechanism manifested exactly: row degrees and pair co-occurrences were uniform, the MM^T identity error was zero, and the predicted signed-membership norm and replication scaling matched observations for both Fano and projective-13 designs. No wall-clock speedup was demonstrated because the Python-loop implementation was slower than dense attention on these tiny cases.", "metrics": { "baseline": "Dense attention: 3,136 score FLOPs and 2.305 ms at n=7; 10,816 FLOPs and 0.298 ms at n=13.", "idea": "Exact design: zero pair-coverage and row-degree variance; signed norm squared matched 4 and 6 for Fano/projective-13; replication sweep norms 4,8,16,32 matched prediction; runtime 15.880 ms at n=7 and 22.086 ms at n=13." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 balanced_design_attention.py", "files": [ "balanced_design_attention.py", "results.json", "REPORT.md" ], "limitations": "No Transformer training, WikiText-2 perplexity, synthetic long-range copy task, fused sparse kernel, large-sequence throughput, or autoregressive KV-cache benchmark was tested. The timing benchmark uses a Python block loop and very small natural design sizes, so it evaluates correctness and routing overhead rather than optimized production performance." }