diff options
| author | YurenHao0426 <blackhao0426@gmail.com> | 2026-06-13 12:35:36 -0500 |
|---|---|---|
| committer | YurenHao0426 <blackhao0426@gmail.com> | 2026-06-13 12:35:36 -0500 |
| commit | 66e0d8b9fd4d0f7a2231d689c055e26fdf1cf04a (patch) | |
| tree | c29cba61124018755a19b02c9d33e3ad5f2e05cc /research/flossing/run_hrm_diag_all_ckpts.sh | |
Curated export for clone-and-run Maze training (2x A6000) + diagnostics.
trm/hrm pretrain.py carry trajectory-augmentation code (backward-compatible).
Heavy artifacts (checkpoints/wandb/npz) gitignored; see PROVENANCE.md.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Diffstat (limited to 'research/flossing/run_hrm_diag_all_ckpts.sh')
| -rwxr-xr-x | research/flossing/run_hrm_diag_all_ckpts.sh | 26 |
1 files changed, 26 insertions, 0 deletions
diff --git a/research/flossing/run_hrm_diag_all_ckpts.sh b/research/flossing/run_hrm_diag_all_ckpts.sh new file mode 100755 index 0000000..e887a5e --- /dev/null +++ b/research/flossing/run_hrm_diag_all_ckpts.sh @@ -0,0 +1,26 @@ +#!/bin/bash +# Diagnose all 10 HRM checkpoints with joint Lyapunov (512 samples each) +set -e +cd /home/yurenh2/rrm/research/flossing +source /home/yurenh2/miniconda3/etc/profile.d/conda.sh +conda activate rrm +export CUDA_VISIBLE_DEVICES=0 + +CKPT_ROOT="/home/yurenh2/rrm/hrm/checkpoints/Sudoku-extreme-1k-aug-1000 ACT-torch/HierarchicalReasoningModel_ACTV1 righteous-python" + +for STEP in step_2604 step_5208 step_7812 step_10416 step_13020 step_15624 step_18228 step_20832 step_23436 step_26040; do + OUT="diag_hrm_${STEP}_512.npz" + LOG="diag_hrm_${STEP}.log" + if [ -f "$OUT" ]; then + echo "skip $STEP (exists)" + continue + fi + echo "[$(date '+%H:%M:%S')] Starting HRM diagnostic $STEP" + python diagnose_hrm_joint.py \ + --ckpt-root "$CKPT_ROOT" --ckpt-name "$STEP" \ + --n-samples 512 --batch-size 32 --k-lyap 8 --t-ons 1 --seed 0 \ + --out "$OUT" > "$LOG" 2>&1 + echo "[$(date '+%H:%M:%S')] $STEP done" +done + +echo "[$(date '+%H:%M:%S')] All HRM diagnostics complete" |