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/multi4_eval_compare/hrm_multi4_eval.csv | |
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/multi4_eval_compare/hrm_multi4_eval.csv')
| -rw-r--r-- | research/flossing/multi4_eval_compare/hrm_multi4_eval.csv | 6 |
1 files changed, 6 insertions, 0 deletions
diff --git a/research/flossing/multi4_eval_compare/hrm_multi4_eval.csv b/research/flossing/multi4_eval_compare/hrm_multi4_eval.csv new file mode 100644 index 0000000..0f7dbc9 --- /dev/null +++ b/research/flossing/multi4_eval_compare/hrm_multi4_eval.csv @@ -0,0 +1,6 @@ +run,step,all/exact_accuracy,all/accuracy,all/lm_loss,all/q_halt_accuracy,all/q_halt_loss,all/steps
+hrm_multi4,2604,0.0123774204403162,0.6303551197052002,0.8603715300559998,0.9879773855209351,0.020903315395116806,16
+hrm_multi4,5208,0.025012653321027756,0.6685170531272888,0.7412638068199158,0.9968069195747375,0.019309692084789276,16
+hrm_multi4,7812,0.04651052877306938,0.687346339225769,0.7044885158538818,0.9902976751327515,0.03646523132920265,16
+hrm_multi4,10416,0.2006617933511734,0.7243355512619019,0.6340938210487366,0.9991532564163208,0.011545917019248009,16
+hrm_multi4,13020,0.34697696566581726,0.7794705033302307,0.49915269017219543,0.9987062215805054,0.01581510715186596,16
|