auto ckpt ckpt_560k.pt

Card: 64M final result (760k, eff 76.07), progress to 64M 760k / 128M 540k, tail-averaging rule, r6 trainer, pool_build, recipe link, board status

Add pool_build/: removed document indices and a standalone script that rebuilds the final training pool (sha256 ecfd0a40) from ARC-MIX

auto ckpt ckpt_540k.pt

Add train_gpt_ref_r6.py, the trainer of the final GoLLeM-v5 runs (sha256 0223c083)

auto ckpt ckpt_520k.pt

auto ckpt ckpt_500k.pt

auto ckpt ckpt_760k.pt

auto ckpt ckpt_480k.pt

auto ckpt ckpt_740k.pt

auto ckpt ckpt_720k.pt

auto ckpt ckpt_460k.pt

auto ckpt ckpt_700k.pt

Card: 64M 640k-680k, 128M 420k/440k (progress rows only)

auto ckpt ckpt_440k.pt

auto ckpt ckpt_680k.pt

auto ckpt ckpt_660k.pt

auto ckpt ckpt_420k.pt

auto ckpt ckpt_640k.pt

Card: 64M 620k, 128M 400k, 400k comparison with v1 128M end of run

auto ckpt ckpt_400k.pt

auto ckpt ckpt_620k.pt

Card: 64M 560k-600k, 128M 380k (progress rows only)

auto ckpt ckpt_600k.pt

auto ckpt ckpt_380k.pt

auto ckpt ckpt_580k.pt

Card: 64M 500k-540k, 128M 340k/360k, v1 128M 360k reference

auto ckpt ckpt_560k.pt

auto ckpt ckpt_360k.pt

auto ckpt ckpt_540k.pt

auto ckpt ckpt_340k.pt

auto ckpt ckpt_520k.pt

auto ckpt ckpt_500k.pt

Card: 64M 440k-480k, 128M 300k/320k, v1 128M 320k reference

auto ckpt ckpt_320k.pt

auto ckpt ckpt_480k.pt

auto ckpt ckpt_300k.pt

auto ckpt ckpt_460k.pt

auto ckpt ckpt_440k.pt

Card: 64M 420k, 128M 280k, v1 128M 280k reference, resume check

auto ckpt ckpt_280k.pt

auto ckpt ckpt_420k.pt

Card: 64M 380k/400k, 128M 260k, 400k comparison with v1 end of run

auto ckpt ckpt_400k.pt

auto ckpt ckpt_260k.pt

auto ckpt ckpt_380k.pt

Card: 128M 240k, v1 128M 240k reference

Card: 64M 340k/360k, 128M 220k, v1 360k reference

auto ckpt ckpt_240k.pt

auto ckpt ckpt_360k.pt