Publish step 5000 checkpoint manifest
Browse files
manifests/checkpoint_step_00005000.json
ADDED
|
@@ -0,0 +1,127 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema_version": 1,
|
| 3 |
+
"status": "complete",
|
| 4 |
+
"config_fingerprint": "0667d4ea7a9eeb08101909c8f4b52860faadd491ce6d4ceb8094405e04a5f75e",
|
| 5 |
+
"dataset_fingerprint": "6cf9eb3d777a338eb007dcc5606b2aac6605bf899a191ea399ccd77d5527acfb",
|
| 6 |
+
"cursor": {
|
| 7 |
+
"global_step": 5000,
|
| 8 |
+
"micro_step": 40124,
|
| 9 |
+
"dataset_epoch": 2,
|
| 10 |
+
"sample_offset": 17600,
|
| 11 |
+
"accumulation_step": 0
|
| 12 |
+
},
|
| 13 |
+
"extra": {
|
| 14 |
+
"last_metrics": {
|
| 15 |
+
"global_step": 5000,
|
| 16 |
+
"behavior_loss": 0.016017724061384797,
|
| 17 |
+
"behavior_microbatches": 8,
|
| 18 |
+
"behavior_singleton_microbatches": 0,
|
| 19 |
+
"oom_replayed_as_singletons": false,
|
| 20 |
+
"gradient_norm": 0.012397946789860725,
|
| 21 |
+
"learning_rate_used": 1e-06,
|
| 22 |
+
"step_seconds": 11.805,
|
| 23 |
+
"peak_reserved_gib": 24.904,
|
| 24 |
+
"optimizer_state_device": "cpu_between_updates",
|
| 25 |
+
"weighted_behavior_objective": 0.016017724061384797,
|
| 26 |
+
"validation_behavior_loss": 0.016737566627562046,
|
| 27 |
+
"lr_window_phase": "collecting",
|
| 28 |
+
"validation_interval": 10,
|
| 29 |
+
"next_validation_step": 5010,
|
| 30 |
+
"learning_rate": 1e-06
|
| 31 |
+
},
|
| 32 |
+
"reason": "session_limit",
|
| 33 |
+
"parent": null,
|
| 34 |
+
"lr_control": {
|
| 35 |
+
"version": 3,
|
| 36 |
+
"config": {
|
| 37 |
+
"factor": 0.5,
|
| 38 |
+
"minimum_learning_rate": 1e-06,
|
| 39 |
+
"window_steps": 50,
|
| 40 |
+
"probe_every_steps": 10,
|
| 41 |
+
"confirmation_steps": 20,
|
| 42 |
+
"minimum_descent_to_noise": 1.0,
|
| 43 |
+
"minimum_relative_descent": 0.0
|
| 44 |
+
},
|
| 45 |
+
"state": {
|
| 46 |
+
"learning_rate": 1e-06,
|
| 47 |
+
"points": [
|
| 48 |
+
{
|
| 49 |
+
"step": 4960,
|
| 50 |
+
"loss": 0.016812349539250135
|
| 51 |
+
},
|
| 52 |
+
{
|
| 53 |
+
"step": 4970,
|
| 54 |
+
"loss": 0.016733078290708362
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"step": 4980,
|
| 58 |
+
"loss": 0.016717635430395603
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"step": 4990,
|
| 62 |
+
"loss": 0.016753521063365043
|
| 63 |
+
},
|
| 64 |
+
{
|
| 65 |
+
"step": 5000,
|
| 66 |
+
"loss": 0.016737566627562046
|
| 67 |
+
}
|
| 68 |
+
],
|
| 69 |
+
"training_points": [
|
| 70 |
+
{
|
| 71 |
+
"step": 4970,
|
| 72 |
+
"loss": 0.015235144022153691
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"step": 4980,
|
| 76 |
+
"loss": 0.013825824274681509
|
| 77 |
+
},
|
| 78 |
+
{
|
| 79 |
+
"step": 4990,
|
| 80 |
+
"loss": 0.013259278662735597
|
| 81 |
+
},
|
| 82 |
+
{
|
| 83 |
+
"step": 5000,
|
| 84 |
+
"loss": 0.013512403922504745
|
| 85 |
+
}
|
| 86 |
+
],
|
| 87 |
+
"training_loss_sum": 0.0,
|
| 88 |
+
"training_loss_count": 0,
|
| 89 |
+
"last_training_step": 5000,
|
| 90 |
+
"window_start_step": 4960,
|
| 91 |
+
"next_validation_step": 5010,
|
| 92 |
+
"last_analysis": {
|
| 93 |
+
"start_step": 4890,
|
| 94 |
+
"end_step": 4960,
|
| 95 |
+
"point_count": 8,
|
| 96 |
+
"log_slope_per_step": 4.04361902428918e-05,
|
| 97 |
+
"fitted_log_descent": -0.0028305333170024257,
|
| 98 |
+
"relative_descent": -0.002834543058775063,
|
| 99 |
+
"residual_mad_scale": 0.0015034014177076266,
|
| 100 |
+
"descent_to_noise": -1.882752858726446,
|
| 101 |
+
"accepted": false,
|
| 102 |
+
"elbow_step": 4890
|
| 103 |
+
},
|
| 104 |
+
"last_training_analysis": {
|
| 105 |
+
"start_step": 4900,
|
| 106 |
+
"end_step": 4960,
|
| 107 |
+
"point_count": 7,
|
| 108 |
+
"log_slope_per_step": -0.00015988120629852886,
|
| 109 |
+
"fitted_log_descent": 0.009592872377911732,
|
| 110 |
+
"relative_descent": 0.009547007553316611,
|
| 111 |
+
"residual_mad_scale": 0.128691405423272,
|
| 112 |
+
"descent_to_noise": 0.07454167080047287,
|
| 113 |
+
"accepted": false,
|
| 114 |
+
"elbow_step": 4900
|
| 115 |
+
},
|
| 116 |
+
"confirming": false,
|
| 117 |
+
"pending_rollback": null,
|
| 118 |
+
"reductions": 9
|
| 119 |
+
}
|
| 120 |
+
},
|
| 121 |
+
"rollback_replay": null
|
| 122 |
+
},
|
| 123 |
+
"files": {
|
| 124 |
+
"adapters.safetensors": "39925cea8660d30a05ffe4b899903351d4c53b80d950efc0e1667837c7a1cb9c",
|
| 125 |
+
"training_state.pt": "61bd503f76b53837fbac2fccf9ba69662dace1d8e39c4b97ff8663703a58ccb2"
|
| 126 |
+
}
|
| 127 |
+
}
|