Auto-sync: 2026-06-26 00:01:50 (part 2)
Browse files
scripts/slurm/eval_h16_rollout.sbatch
CHANGED
|
@@ -29,7 +29,7 @@ CACHE_DIR="/tmp/$USER/dovla-mesa-$SLURM_JOB_ID"
|
|
| 29 |
|
| 30 |
SEED=$SLURM_ARRAY_TASK_ID
|
| 31 |
CHECKPOINT="$SCRATCH_ROOT/experiments/h16_policy_runs/seed_$SEED/best.pt"
|
| 32 |
-
DATASET="$SCRATCH_ROOT/experiments/
|
| 33 |
OUT="$SCRATCH_ROOT/experiments/h16_policy_runs/seed_${SEED}/online_rollout.json"
|
| 34 |
|
| 35 |
module load StdEnv/2023 apptainer/1.4.5
|
|
|
|
| 29 |
|
| 30 |
SEED=$SLURM_ARRAY_TASK_ID
|
| 31 |
CHECKPOINT="$SCRATCH_ROOT/experiments/h16_policy_runs/seed_$SEED/best.pt"
|
| 32 |
+
DATASET="$SCRATCH_ROOT/experiments/h16_merged_dataset"
|
| 33 |
OUT="$SCRATCH_ROOT/experiments/h16_policy_runs/seed_${SEED}/online_rollout.json"
|
| 34 |
|
| 35 |
module load StdEnv/2023 apptainer/1.4.5
|
scripts/sync_training_outputs.py
CHANGED
|
@@ -10,7 +10,7 @@ from pathlib import Path
|
|
| 10 |
from huggingface_hub import HfApi, upload_file, upload_folder
|
| 11 |
|
| 12 |
REPO_ID = "anhtld/vla"
|
| 13 |
-
TRAINING_JOBS = [
|
| 14 |
CHECK_INTERVAL = 300 # 5 minutes
|
| 15 |
SCRATCH = Path("/scratch/knguy52/dovla/experiments")
|
| 16 |
|
|
|
|
| 10 |
from huggingface_hub import HfApi, upload_file, upload_folder
|
| 11 |
|
| 12 |
REPO_ID = "anhtld/vla"
|
| 13 |
+
TRAINING_JOBS = [14758888] # h=16 evaluation job
|
| 14 |
CHECK_INTERVAL = 300 # 5 minutes
|
| 15 |
SCRATCH = Path("/scratch/knguy52/dovla/experiments")
|
| 16 |
|