| { |
| "run": "G3A1 joint R2R + English RxR15, one-epoch decay (3,852 updates, warmup 116, cosine to 0.1 x 5e-6)", |
| "global_step": 2100, |
| "max_steps": 3852, |
| "epoch": 0.5451713395638629, |
| "percent": 54.52, |
| "remaining_updates": 1752, |
| "learning_rate_at_stop": 2.5327567618061965e-06, |
| "loss_at_stop": 0.1690405160188675, |
| "grad_norm_at_stop": 2.477595806121826, |
| "mean_loss_last_100_updates": 0.15498921513557434, |
| "aux_weight_at_stop": 0.1, |
| "g3_state_loss_per_pair_last_100": 0.013914937848588878, |
| "a1_relation_loss_per_pair_last_100": 0.3475379606630543, |
| "exposures": 16800, |
| "unique_episodes_seen": 16800, |
| "duplicate_exposures": 0, |
| "manifest_episodes": 30815, |
| "epoch_exposures": 30816, |
| "remaining_exposures": 14016, |
| "actions_seen": 1708373, |
| "manifest_actions": 3128624, |
| "sampler": "EpisodeLengthSampler seed 429 (position restored from checkpoint trainer_state/RNG; do not change the manifest order)", |
| "cuda_ooms": 0, |
| "allocations": [ |
| { |
| "job": "926610", |
| "node": "evc104 (4 CPUs, 512G)", |
| "updates": "1-501", |
| "stop": "deadline checkpoint-stop" |
| }, |
| { |
| "job": "930602", |
| "node": "evc101 (8 CPUs, 768G)", |
| "updates": "502-2100", |
| "stop": "user-requested clean stop at 2100" |
| } |
| ], |
| "resume_from": "checkpoint-2100 (complete, 4 ranks)" |
| } |
|
|
