summaryrefslogtreecommitdiff
path: root/scripts
diff options
context:
space:
mode:
authoryurenh <blackhao0426@gmail.com>2026-08-31 18:26:39 -0500
committeryurenh <blackhao0426@gmail.com>2026-08-31 18:26:39 -0500
commit5e5f56cff99fe017c08d56d1303773d1299c0e38 (patch)
tree56e100f921a93717aa3936eafd547f0cecbeddb5 /scripts
parent76652293b39fd97a2e7870e46ad3af63f9457cc1 (diff)
ladder runner: SET override passthrough; validated end-to-end
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GkgLsACEF6CCP7EUfA5fZe
Diffstat (limited to 'scripts')
-rwxr-xr-xscripts/run_ladder.sh4
1 files changed, 3 insertions, 1 deletions
diff --git a/scripts/run_ladder.sh b/scripts/run_ladder.sh
index 764632a..65b43a9 100755
--- a/scripts/run_ladder.sh
+++ b/scripts/run_ladder.sh
@@ -10,6 +10,7 @@
# PREP_TOKENS=3e9 tokens to tokenize if DATA missing [3e9]
# OUT=runs output root [runs]
# 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
#
# Runs are sequential (each takes the whole node), resume-safe: a finished run leaves OUT/<name>/DONE
@@ -24,7 +25,8 @@ DATA=${DATA:-data/fineweb}
PREP_TOKENS=${PREP_TOKENS:-3e9}
OUT=${OUT:-runs}
MICRO_BS=${MICRO_BS:-8}
-EXTRA=${TOKENS:+--set tokens=$TOKENS}
+EXTRA="--set ${TOKENS:+tokens=$TOKENS} ${SET:-}"
+[ "$EXTRA" = "--set " ] && EXTRA=""
echo "== zbp-scaling ladder: sizes=[$SIZES] arms=[$ARMS] nproc=$NPROC data=$DATA out=$OUT ${TOKENS:+tokens=$TOKENS}"
[ "$NPROC" -ge 1 ] || { echo "no GPUs detected; set NPROC"; exit 1; }