vibego-distilled-b18-kata1

Recent kata1 training positions (daily archives 2026-05-04..06, multi-board-size, randomized komi) relabeled by b18 at visits 1 — the study's base distillation set. ~44M positions, 5379 shards, ~75GB. Champion nets trained on this + the g170mix set.

Format

Each .npz holds training rows in KataGo-v7-compatible arrays:

• binaryInputNCHWPacked (N, 22, 46) uint8 — packed 19x19 binary input planes (v7 channel order)
• globalInputNC (N, 19) float32 — global features (komi at index 5 = selfKomi/20, side-to-move relative)
• policyTargetsNCMove (N, 1, 362) float32 — teacher policy target (361 moves + pass), to-move
• globalTargetsNC (N, 64) float32 — [0:3] win/loss/nr, [3] scoreLead (to-move), [27] ownership weight
• valueTargetsNCHW (N, 1, 19, 19) — ownership target, to-move-positive

Teacher: public kata1-b18c384nbt queried at the stated visits. Upstream positions from katagoarchive.org (g170 self-play / kata1 daily training data) — credit to lightvector and the KataGo distributed-training contributors. License: CC-BY-4.0.