fam/nanochat-omni
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

at 28 and above we start to need batch size 8

Browse Source
This commit is contained in:
Andrej Karpathy
2026-02-08 18:26:34 +00:00
parent ff46300720
commit 1ec0a34779
1 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
runs/miniseries.sh
+4 -2
View File
@@ -57,8 +57,10 @@ for d in "${DEPTHS[@]}"; do
TAG="${SERIES_NAME}_miniseries_d${d}" TAG="${SERIES_NAME}_miniseries_d${d}"
START_TIME=$(date +%s) START_TIME=$(date +%s)
# For depths >= 22, use smaller device batch size to avoid OOM # Reduce --device-batch-size to avoid OOM at larger depths
if [ $d -ge 22 ]; then if [ $d -ge 28 ]; then
DEVICE_BATCH_SIZE_ARG="--device-batch-size=8"
elif [ $d -ge 20 ]; then
DEVICE_BATCH_SIZE_ARG="--device-batch-size=16" DEVICE_BATCH_SIZE_ARG="--device-batch-size=16"
else else
DEVICE_BATCH_SIZE_ARG="--device-batch-size=32" DEVICE_BATCH_SIZE_ARG="--device-batch-size=32"