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/smoke_scratch_trm_baseline.json | |
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/smoke_scratch_trm_baseline.json')
| -rw-r--r-- | research/flossing/smoke_scratch_trm_baseline.json | 72 |
1 files changed, 72 insertions, 0 deletions
diff --git a/research/flossing/smoke_scratch_trm_baseline.json b/research/flossing/smoke_scratch_trm_baseline.json new file mode 100644 index 0000000..cc6e926 --- /dev/null +++ b/research/flossing/smoke_scratch_trm_baseline.json @@ -0,0 +1,72 @@ +{ + "args": { + "model": "trm", + "ckpt_root": "/home/yurenh2/rrm/trm/checkpoints/Sudoku-extreme-1k-aug-1000-ACT-torch/pretrain_mlp_t_sudoku_singleGPU", + "ckpt_name": null, + "init_from": "random", + "mode": "baseline_clean", + "train_steps": 1, + "batch_size": 2, + "lr": 0.0001, + "noise_std": 0.001, + "noise_min": null, + "noise_max": null, + "noise_sampling": "normal", + "clean_prob": 0.0, + "sigma_start": null, + "sigma_ramp_steps": 0, + "n_trajectories": 4, + "rollout_impl": "parallel_fixed", + "perturb": "both", + "seed": 2027, + "eval_every": 1, + "eval_n": 64, + "eval_batch_size": 16, + "out": "smoke_scratch_trm_baseline.json", + "save_dir": null, + "save_best": false, + "save_final": false, + "save_every_eval": false, + "save_train_state": false, + "resume_state": null + }, + "initial_acc": 0.0, + "initial_tok_acc": 0.0933641975308642, + "steps": [ + { + "train_step": 1, + "loss": 2.542586088180542, + "clean_loss": 2.542586088180542, + "noisy_loss_mean": 0.0, + "noise_std_target": 0.001, + "noise_std_mean": 0.0, + "noise_std_max": 0.0, + "effective_batch": 2 + } + ], + "evals": [ + { + "kind": "initial", + "train_step": 0, + "acc": 0.0, + "tok_acc": 0.0933641975308642 + }, + { + "kind": "task", + "train_step": 1, + "acc": 0.0, + "tok_acc": 0.09510030864197531 + }, + { + "kind": "final", + "train_step": 1, + "acc": 0.0, + "tok_acc": 0.09510030864197531 + } + ], + "checkpoints": [], + "resume_state": null, + "resume_step": 0, + "final_acc": 0.0, + "final_tok_acc": 0.09510030864197531 +}
\ No newline at end of file |