Publish step 3510 checkpoint manifest
Browse files
manifests/checkpoint_step_00003510.json
ADDED
|
@@ -0,0 +1,115 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"status": "complete",
|
| 4 |
+
"config_fingerprint": "0667d4ea7a9eeb08101909c8f4b52860faadd491ce6d4ceb8094405e04a5f75e",
|
| 5 |
+
"dataset_fingerprint": "6cf9eb3d777a338eb007dcc5606b2aac6605bf899a191ea399ccd77d5527acfb",
|
| 6 |
+
"cursor": {
|
| 7 |
+
"global_step": 3510,
|
| 8 |
+
"micro_step": 28165,
|
| 9 |
+
"dataset_epoch": 1,
|
| 10 |
+
"sample_offset": 24960,
|
| 11 |
+
"accumulation_step": 0
|
| 12 |
+
},
|
| 13 |
+
"extra": {
|
| 14 |
+
"last_metrics": {
|
| 15 |
+
"global_step": 3510,
|
| 16 |
+
"behavior_loss": 0.014460635284194723,
|
| 17 |
+
"behavior_microbatches": 8,
|
| 18 |
+
"behavior_singleton_microbatches": 0,
|
| 19 |
+
"oom_replayed_as_singletons": false,
|
| 20 |
+
"gradient_norm": 0.022714726626873016,
|
| 21 |
+
"learning_rate_used": 1e-06,
|
| 22 |
+
"step_seconds": 12.606,
|
| 23 |
+
"peak_reserved_gib": 24.904,
|
| 24 |
+
"optimizer_state_device": "cpu_between_updates",
|
| 25 |
+
"weighted_behavior_objective": 0.014460635284194723,
|
| 26 |
+
"validation_behavior_loss": 0.01682164315134287,
|
| 27 |
+
"lr_window_phase": "minimum_lr_plateau",
|
| 28 |
+
"lr_window_start_step": 3440,
|
| 29 |
+
"lr_window_end_step": 3510,
|
| 30 |
+
"lr_window_point_count": 8,
|
| 31 |
+
"lr_window_log_slope_per_step": 3.468046811199166e-05,
|
| 32 |
+
"lr_window_fitted_log_descent": -0.002427632767839416,
|
| 33 |
+
"lr_window_relative_descent": -0.002430581854217184,
|
| 34 |
+
"lr_window_residual_mad_scale": 0.0030586733246124,
|
| 35 |
+
"lr_window_descent_to_noise": -0.7936881484874001,
|
| 36 |
+
"lr_window_accepted": false,
|
| 37 |
+
"lr_window_elbow_step": 3450,
|
| 38 |
+
"lr_training_window_start_step": 3450,
|
| 39 |
+
"lr_training_window_end_step": 3510,
|
| 40 |
+
"lr_training_window_point_count": 7,
|
| 41 |
+
"lr_training_window_log_slope_per_step": 0.002190928189992114,
|
| 42 |
+
"lr_training_window_fitted_log_descent": -0.13145569139952684,
|
| 43 |
+
"lr_training_window_relative_descent": -0.1404873732034586,
|
| 44 |
+
"lr_training_window_residual_mad_scale": 0.012542356750465129,
|
| 45 |
+
"lr_training_window_descent_to_noise": -10.48094022637746,
|
| 46 |
+
"lr_training_window_accepted": false,
|
| 47 |
+
"lr_training_window_elbow_step": 3460,
|
| 48 |
+
"lr_window_any_signal_accepted": false,
|
| 49 |
+
"validation_interval": 10,
|
| 50 |
+
"next_validation_step": 3520,
|
| 51 |
+
"learning_rate": 1e-06
|
| 52 |
+
},
|
| 53 |
+
"reason": "periodic",
|
| 54 |
+
"parent": null,
|
| 55 |
+
"lr_control": {
|
| 56 |
+
"version": 3,
|
| 57 |
+
"config": {
|
| 58 |
+
"factor": 0.5,
|
| 59 |
+
"minimum_learning_rate": 1e-06,
|
| 60 |
+
"window_steps": 50,
|
| 61 |
+
"probe_every_steps": 10,
|
| 62 |
+
"confirmation_steps": 20,
|
| 63 |
+
"minimum_descent_to_noise": 1.0,
|
| 64 |
+
"minimum_relative_descent": 0.0
|
| 65 |
+
},
|
| 66 |
+
"state": {
|
| 67 |
+
"learning_rate": 1e-06,
|
| 68 |
+
"points": [
|
| 69 |
+
{
|
| 70 |
+
"step": 3510,
|
| 71 |
+
"loss": 0.01682164315134287
|
| 72 |
+
}
|
| 73 |
+
],
|
| 74 |
+
"training_points": [],
|
| 75 |
+
"training_loss_sum": 0.0,
|
| 76 |
+
"training_loss_count": 0,
|
| 77 |
+
"last_training_step": 3510,
|
| 78 |
+
"window_start_step": 3510,
|
| 79 |
+
"next_validation_step": 3520,
|
| 80 |
+
"last_analysis": {
|
| 81 |
+
"start_step": 3440,
|
| 82 |
+
"end_step": 3510,
|
| 83 |
+
"point_count": 8,
|
| 84 |
+
"log_slope_per_step": 3.468046811199166e-05,
|
| 85 |
+
"fitted_log_descent": -0.002427632767839416,
|
| 86 |
+
"relative_descent": -0.002430581854217184,
|
| 87 |
+
"residual_mad_scale": 0.0030586733246124,
|
| 88 |
+
"descent_to_noise": -0.7936881484874001,
|
| 89 |
+
"accepted": false,
|
| 90 |
+
"elbow_step": 3450
|
| 91 |
+
},
|
| 92 |
+
"last_training_analysis": {
|
| 93 |
+
"start_step": 3450,
|
| 94 |
+
"end_step": 3510,
|
| 95 |
+
"point_count": 7,
|
| 96 |
+
"log_slope_per_step": 0.002190928189992114,
|
| 97 |
+
"fitted_log_descent": -0.13145569139952684,
|
| 98 |
+
"relative_descent": -0.1404873732034586,
|
| 99 |
+
"residual_mad_scale": 0.012542356750465129,
|
| 100 |
+
"descent_to_noise": -10.48094022637746,
|
| 101 |
+
"accepted": false,
|
| 102 |
+
"elbow_step": 3460
|
| 103 |
+
},
|
| 104 |
+
"confirming": false,
|
| 105 |
+
"pending_rollback": null,
|
| 106 |
+
"reductions": 9
|
| 107 |
+
}
|
| 108 |
+
},
|
| 109 |
+
"rollback_replay": null
|
| 110 |
+
},
|
| 111 |
+
"files": {
|
| 112 |
+
"adapters.safetensors": "3f78038fb2a496d336c5437a1129ce105ca66448e2de752e8f1abe8650c74700",
|
| 113 |
+
"training_state.pt": "62aa73c036a6c23d5b51eba117eeaf32582b72c87c27309a2e0a424a52730921"
|
| 114 |
+
}
|
| 115 |
+
}
|