perdim trainmax CTT 2026-07-03 workspace/scripts/slurm/eval_ctt_generated_rollout.sbatch
Browse files
workspace/scripts/slurm/eval_ctt_generated_rollout.sbatch
CHANGED
|
@@ -13,10 +13,24 @@
|
|
| 13 |
set -euo pipefail
|
| 14 |
|
| 15 |
PROJECT_DIR="${PROJECT_DIR:-$SLURM_SUBMIT_DIR}"
|
| 16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
SOURCE_INDEX="${SOURCE_INDEX:-data/cil_charts/train/index.json}"
|
| 18 |
TARGET_INDEX="${TARGET_INDEX:-data/cil_charts/val/index.json}"
|
| 19 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
|
| 21 |
SCRATCH_ROOT="/scratch/$USER/dovla"
|
| 22 |
SIF="${SIF:-$SCRATCH_ROOT/containers/pytorch_2.7.1_cuda12.8.sif}"
|
|
|
|
| 13 |
set -euo pipefail
|
| 14 |
|
| 15 |
PROJECT_DIR="${PROJECT_DIR:-$SLURM_SUBMIT_DIR}"
|
| 16 |
+
TASK_ID="${SLURM_ARRAY_TASK_ID:-0}"
|
| 17 |
+
if [[ -z "${CHECKPOINT:-}" ]]; then
|
| 18 |
+
if [[ -n "${CHECKPOINT_TEMPLATE:-}" ]]; then
|
| 19 |
+
CHECKPOINT="${CHECKPOINT_TEMPLATE//\{seed\}/$TASK_ID}"
|
| 20 |
+
else
|
| 21 |
+
echo "Set CHECKPOINT to a CTT model.pt or CHECKPOINT_TEMPLATE with {seed}" >&2
|
| 22 |
+
exit 2
|
| 23 |
+
fi
|
| 24 |
+
fi
|
| 25 |
SOURCE_INDEX="${SOURCE_INDEX:-data/cil_charts/train/index.json}"
|
| 26 |
TARGET_INDEX="${TARGET_INDEX:-data/cil_charts/val/index.json}"
|
| 27 |
+
if [[ -z "${OUT_DIR:-}" ]]; then
|
| 28 |
+
if [[ -n "${OUT_DIR_TEMPLATE:-}" ]]; then
|
| 29 |
+
OUT_DIR="${OUT_DIR_TEMPLATE//\{seed\}/$TASK_ID}"
|
| 30 |
+
else
|
| 31 |
+
OUT_DIR="runs/ctt_residual_rollout_smoke_seed${TASK_ID}"
|
| 32 |
+
fi
|
| 33 |
+
fi
|
| 34 |
|
| 35 |
SCRATCH_ROOT="/scratch/$USER/dovla"
|
| 36 |
SIF="${SIF:-$SCRATCH_ROOT/containers/pytorch_2.7.1_cuda12.8.sif}"
|