From 66e0d8b9fd4d0f7a2231d689c055e26fdf1cf04a Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Sat, 13 Jun 2026 12:35:36 -0500 Subject: rrm workspace: TRM/HRM/SRM code, Maze dataset, dynamical-analysis pipeline 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 --- .../initial_perturb_robustness/smoke_baseline_b32k8.summary.csv | 2 ++ 1 file changed, 2 insertions(+) create mode 100644 research/flossing/initial_perturb_robustness/smoke_baseline_b32k8.summary.csv (limited to 'research/flossing/initial_perturb_robustness/smoke_baseline_b32k8.summary.csv') diff --git a/research/flossing/initial_perturb_robustness/smoke_baseline_b32k8.summary.csv b/research/flossing/initial_perturb_robustness/smoke_baseline_b32k8.summary.csv new file mode 100644 index 0000000..e7ca899 --- /dev/null +++ b/research/flossing/initial_perturb_robustness/smoke_baseline_b32k8.summary.csv @@ -0,0 +1,2 @@ +label,sigma,n_samples,rollouts,ckpt_root,ckpt_name,perturb,noise_distribution,mean_rollout_exact,mean_rollout_token_acc,pass_at_k,all_k,correct_count_mean,correct_count_std,correct_count_q10,correct_count_q50,correct_count_q90,zero_frac,full_frac +trm_baseline_b32k8_smoke,0.0,32,8,/home/yurenh2/rrm/trm/checkpoints/Sudoku-extreme-1k-aug-1000-ACT-torch/pretrain_mlp_t_sudoku_official_gbs768_repro,step_58590,both,gaussian,0.90625,0.9637345671653748,0.90625,0.90625,7.25,2.3318448066711426,8.0,8.0,8.0,0.09375,0.90625 -- cgit v1.2.3