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 --- research/flossing/multi4_eval_compare/trm_baseline_eval.csv | 11 +++++++++++ 1 file changed, 11 insertions(+) create mode 100644 research/flossing/multi4_eval_compare/trm_baseline_eval.csv (limited to 'research/flossing/multi4_eval_compare/trm_baseline_eval.csv') diff --git a/research/flossing/multi4_eval_compare/trm_baseline_eval.csv b/research/flossing/multi4_eval_compare/trm_baseline_eval.csv new file mode 100644 index 0000000..0b6bc11 --- /dev/null +++ b/research/flossing/multi4_eval_compare/trm_baseline_eval.csv @@ -0,0 +1,11 @@ +run,step,all/exact_accuracy,all/accuracy,all/lm_loss,all/q_halt_accuracy,all/q_halt_loss,all/steps +trm_baseline,26041,0.5575894117355347,0.8469749093055725,0.35285070538520813,0.9997445344924927,0.003797376062721014,16 +trm_baseline,52082,0.6295099854469299,0.8704391121864319,0.2973524034023285,0.9998486042022705,0.0011324305087327957,16 +trm_baseline,78123,0.6993892788887024,0.8920264840126038,0.2477506548166275,0.9998935461044312,0.0006260558147914708,16 +trm_baseline,104164,0.72928386926651,0.9020143151283264,0.22608688473701477,0.9998770356178284,0.001141014276072383,16 +trm_baseline,130205,0.7653990387916565,0.914251446723938,0.19878524541854858,0.999917209148407,0.0010435190051794052,16 +trm_baseline,156246,0.7596656680107117,0.9119072556495667,0.2041437327861786,0.999862790107727,0.0011596218682825565,16 +trm_baseline,182287,0.7541900873184204,0.9094774723052979,0.2100999504327774,0.9998249411582947,0.0013093978632241488,16 +trm_baseline,208328,0.7732800841331482,0.9166732430458069,0.19410833716392517,0.9998320937156677,0.0033004307188093662,16 +trm_baseline,234369,0.7750374674797058,0.9172152280807495,0.19279460608959198,0.9998533725738525,0.0032994903158396482,16 +trm_baseline,260410,0.7742025256156921,0.9169425964355469,0.19348150491714478,0.9998462796211243,0.002894919365644455,16 -- cgit v1.2.3