For Minegishi: training documentation, reproducibility files and line plots 4/4

For Minegishi: training documentation, reproducibility files and line plots 3/4

For Minegishi: training documentation, reproducibility files and line plots 2/4

For Minegishi: training documentation, reproducibility files and line plots 1/4

For Minegishi: RoPE10000 seed7 checkpoints 45-46 of 46

For Minegishi: RoPE10000 seed7 checkpoints 41-44 of 46

For Minegishi: RoPE10000 seed7 checkpoints 37-40 of 46

For Minegishi: RoPE10000 seed7 checkpoints 33-36 of 46

For Minegishi: RoPE10000 seed7 checkpoints 29-32 of 46

For Minegishi: RoPE10000 seed7 checkpoints 25-28 of 46

For Minegishi: RoPE10000 seed7 checkpoints 21-24 of 46

For Minegishi: RoPE10000 seed7 checkpoints 17-20 of 46

For Minegishi: RoPE10000 seed7 checkpoints 13-16 of 46

For Minegishi: RoPE10000 seed7 checkpoints 9-12 of 46

For Minegishi: RoPE10000 seed7 checkpoints 5-8 of 46

For Minegishi: RoPE10000 seed7 checkpoints 1-4 of 46

data/chain_loop (k = 1 default data dir) and README section on resuming

Resume states (last.pt with optimizer and stream state) of the runs that were queued at the 2026-09-25 maintenance stop, plus the five data directories

Loop transformer halting line, 2026-09-24 snapshot: 79 runs, 91 model checkpoints, evaluations, probes

add rl-mda-step100

add rl-mda-step50

add rl-mda-step5

add rl-r1b-step100

add sft-matched-grouped-s1

add sft-matched-isolated-s1

add sft-c4b-s1

add sft-c4-s1

add stage15b-init

README: TL;DR with pairwise contrasts (fixed nested code fences)

add README

add rl-r1a-step100

add rl-mdb-step100

add sft-d1-s1

add sft-c1-s1

add sft-v1-s1

add README

add eval files

add sft-eco-s1

initial commit