diff options
Diffstat (limited to 'scripts/run_ladder.sh')
| -rwxr-xr-x | scripts/run_ladder.sh | 15 |
1 files changed, 14 insertions, 1 deletions
diff --git a/scripts/run_ladder.sh b/scripts/run_ladder.sh index 1d412a0..7402111 100755 --- a/scripts/run_ladder.sh +++ b/scripts/run_ladder.sh @@ -12,6 +12,8 @@ # MICRO_BS=8 per-GPU micro batch [8] # SET="k=v k2=v2" extra --set overrides for every run (e.g. vocab=8192 seq_len=256) # DRY=1 print the plan and exit +# PUSH_RESULTS=0 skip the auto collect + git-push of results JSONL [on] +# HF_UPLOAD=1 [HF_CKPT=1] also upload to a private HF repo (needs HF_TOKEN; see README) # # Runs are sequential (each takes the whole node), resume-safe: a finished run leaves OUT/<name>/DONE # and is skipped on re-invocation, so the script can be re-run after interruptions. @@ -66,8 +68,19 @@ for size in $SIZES; do touch "$dir/DONE" done done +TAG=${HF_TAG:-$(hostname)-$(date +%Y%m%d)} +if [ "${PUSH_RESULTS:-1}" = 1 ] && [ "${DRY:-0}" != 1 ]; then + # zero-token default: collect the small JSONL results and push them back over this clone's git auth + if python scripts/collect.py --runs "$OUT" --out "results/$TAG"; then + git add results + git -c user.email=ladder@zbp -c user.name=ladder commit -m "results: $TAG" || true # nothing new is fine + git push || echo "!! git push failed — push manually later or send results/$TAG" + else + echo "!! collect found no finished runs — skipping push" + fi +fi if [ "${HF_UPLOAD:-0}" = 1 ]; then - TAG=${HF_TAG:-$(hostname)-$(date +%Y%m%d)} + # optional direct-to-HF (needs HF_TOKEN in the environment; see README Security) python scripts/collect.py --runs "$OUT" --out "results/$TAG" python scripts/upload_hf.py --results "results/$TAG" ${HF_REPO:+--repo "$HF_REPO"} ${HF_CKPT:+--with-ckpt "$OUT"} fi |