| Age | Commit message (Expand) | Author |
|---|---|---|
| 2026-09-19 | transposing the weights gives ~2.5x speedup | tslil |
| 2026-09-19 | big refactor | tslil |
| 2026-09-18 | working towards hybrid rootstrap | tslil |
| 2026-09-13 | move data generation into the epoch loop | tslil |
| 2026-09-12 | still observing non-monotonic improvements between training generations | tslil |
| 2026-09-12 | flood fill for roads, fix dragon, memoise leaf eval, enable diagonal in nn_tr... | tslil |
| 2026-09-11 | move from single candidate td-leaf with gating to tournament of a population | tslil |
| 2026-09-10 | credit Mitchell Riley: we can drop 65 parameters for about the same strength;... | tslil |
| 2026-09-10 | use silu instead of relu with the intuition being that we avoid dead neurons | tslil |
| 2026-09-05 | training rework, best weights so far* | tslil |
| 2026-09-04 | TD-leaf and another negamax bug | tslil |
| 2026-09-02 | TD(lambda), and a big rework | tslil |
| 2023-01-21 | might as well enable size 6 | tslil clingman |
