diff options
Diffstat (limited to 'research/flossing/launch_trajectory_perturb_queue.sh')
| -rwxr-xr-x | research/flossing/launch_trajectory_perturb_queue.sh | 92 |
1 files changed, 92 insertions, 0 deletions
diff --git a/research/flossing/launch_trajectory_perturb_queue.sh b/research/flossing/launch_trajectory_perturb_queue.sh new file mode 100755 index 0000000..ba681ed --- /dev/null +++ b/research/flossing/launch_trajectory_perturb_queue.sh @@ -0,0 +1,92 @@ +#!/usr/bin/env bash +set -eo pipefail + +MODEL="${1:-hrm}" +GPU="${2:-0}" + +source /home/yurenh2/miniconda3/etc/profile.d/conda.sh +conda activate rrm +cd /home/yurenh2/rrm/research/flossing + +HRM_ROOT="/home/yurenh2/rrm/hrm/checkpoints/Sudoku-extreme-1k-aug-1000 ACT-torch/HierarchicalReasoningModel_ACTV1 righteous-python" +TRM_ROOT="/home/yurenh2/rrm/trm/checkpoints/Sudoku-extreme-1k-aug-1000-ACT-torch/pretrain_mlp_t_sudoku_singleGPU" + +run_one() { + local tag="$1" + shift + echo "[$(date -Is)] START ${tag}" + CUDA_VISIBLE_DEVICES="${GPU}" python step9_trajectory_perturb_train.py "$@" \ + > "${tag}.log" 2>&1 + echo "[$(date -Is)] DONE ${tag}" +} + +if [[ "${MODEL}" == "hrm" ]]; then + run_one step9_A_hrm_single_perturb_sigma1e-3_26040_10k \ + --model hrm \ + --ckpt-root "${HRM_ROOT}" \ + --ckpt-name step_26040 \ + --mode single_perturbed_ce \ + --train-steps 10000 \ + --batch-size 8 \ + --lr 1e-5 \ + --noise-std 0.001 \ + --perturb both \ + --seed 42 \ + --eval-every 1000 \ + --eval-n 512 \ + --eval-batch-size 32 \ + --out step9_A_hrm_single_perturb_sigma1e-3_26040_10k.json + + run_one step9_B_hrm_multi4_perturb_sigma1e-3_26040_10k \ + --model hrm \ + --ckpt-root "${HRM_ROOT}" \ + --ckpt-name step_26040 \ + --mode multi_perturbed_ce \ + --n-trajectories 4 \ + --train-steps 10000 \ + --batch-size 8 \ + --lr 1e-5 \ + --noise-std 0.001 \ + --perturb both \ + --seed 42 \ + --eval-every 1000 \ + --eval-n 512 \ + --eval-batch-size 32 \ + --out step9_B_hrm_multi4_perturb_sigma1e-3_26040_10k.json +elif [[ "${MODEL}" == "trm" ]]; then + run_one step9_C_trm_single_perturb_sigma1e-3_26041_batch4_10k \ + --model trm \ + --ckpt-root "${TRM_ROOT}" \ + --ckpt-name step_26041 \ + --mode single_perturbed_ce \ + --train-steps 10000 \ + --batch-size 4 \ + --lr 1e-5 \ + --noise-std 0.001 \ + --perturb both \ + --seed 42 \ + --eval-every 1000 \ + --eval-n 512 \ + --eval-batch-size 32 \ + --out step9_C_trm_single_perturb_sigma1e-3_26041_batch4_10k.json + + run_one step9_D_trm_multi4_perturb_sigma1e-3_26041_batch4_10k \ + --model trm \ + --ckpt-root "${TRM_ROOT}" \ + --ckpt-name step_26041 \ + --mode multi_perturbed_ce \ + --n-trajectories 4 \ + --train-steps 10000 \ + --batch-size 4 \ + --lr 1e-5 \ + --noise-std 0.001 \ + --perturb both \ + --seed 42 \ + --eval-every 1000 \ + --eval-n 512 \ + --eval-batch-size 32 \ + --out step9_D_trm_multi4_perturb_sigma1e-3_26041_batch4_10k.json +else + echo "unknown model: ${MODEL}" >&2 + exit 2 +fi |
