mirror of
https://github.com/hanzoai/benchmarks.git
synced 2026-10-30 14:59:11 +00:00
zeekay
1b9ddf36dd
cardinality.json carries kai-1's keys (weights 0834a74f, bench cap on CUDA in F32, 2026-10-03) beside Laya's and Jev's, unchanged, and results/cardinality/kai.* its answers and summaries at every size. Every option scored: 4 options 96.75%, 16 93.50%, 77 90.75%, 150 90.75%, against Jev's 90.50%, 85.25%, 84.25% and 94.00%. cardinality.py builds the cases at 4, 16, 77 and 150 options as bench cap does, asks Kai over POST /v1/decisions, and prints its accuracy beside the recorded kai-1 and Jev. --check rebuilds the cases and reproduces both records question by question. Jev is never called.