Add immutable checkpoint backup cosmos3_nano_stage_one_lora_r256_replicate_only_500k_step_008000_20261105 run_execution_metadata
Browse files
cosmos3_nano_stage_one_lora_r256_replicate_only_500k_20261103/attempt01/step_008000/backup_context.json
ADDED
|
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"all_finite": true,
|
| 3 |
+
"base_model": "nvidia/Cosmos3-Nano",
|
| 4 |
+
"base_revision": "e59a53c25979a090fa8706c9acc0c254a6e89b92",
|
| 5 |
+
"checkpoint_kind": "trainable_parameters_only_safetensors",
|
| 6 |
+
"checkpoint_sha256": "315150ff1ad5354c3fc92760058b68b711016ed54c3ffd810c4c8dc0d46d6918",
|
| 7 |
+
"checkpoint_source": "/scratch/shared/engs2907/logs/vgg_forecast_4d/train/cosmos3_nano_five_stream_stage_one_lora_r256_replicate_only_500k_20261103/attempt04/trainable_overlays/step_008000.safetensors",
|
| 8 |
+
"config_identity": "cosmos3_nano_omni4d_stage_one_lora_r256_native_layer_masks_fp32_rope_full_checkpoint_retained_forward_parameters_global32",
|
| 9 |
+
"dtype": "float32",
|
| 10 |
+
"exclusions": [
|
| 11 |
+
"pretrained base weights",
|
| 12 |
+
"optimizer state",
|
| 13 |
+
"full model checkpoint"
|
| 14 |
+
],
|
| 15 |
+
"file_bytes_expected": 2990807184,
|
| 16 |
+
"lineage": {
|
| 17 |
+
"exact_training_resume": false,
|
| 18 |
+
"optimizer_state_included": false,
|
| 19 |
+
"parent_checkpoint_path": "/scratch/shared/engs2907/logs/vgg_forecast_4d/tools/hf_checkpoint_backup/cosmos3_nano_stage_one_lora_r256_replicate_only_500k_20261104/attempt06/update_007300/run/checkpoints/adapters/step_007300.safetensors",
|
| 20 |
+
"parent_checkpoint_sha256": "9cbbeb3c9e01ca0f04f7ac35d1fd09b1c6f4dd1196bd764f1e75254fc88846a5",
|
| 21 |
+
"parent_update": 7300,
|
| 22 |
+
"pretrained_base_included": false
|
| 23 |
+
},
|
| 24 |
+
"model": "Cosmos3-Nano Omni4D stage one",
|
| 25 |
+
"parameter_count": 747673344,
|
| 26 |
+
"resolved_config_path": "/scratch/shared/engs2907/logs/vgg_forecast_4d/train/cosmos3/omni4d_stage_one_formal/cosmos3_nano_five_stream_stage_one_lora_r256_replicate_only_500k_20261103_attempt04/config.yaml",
|
| 27 |
+
"resolved_config_sha256": "f7bb637431ad417f9208a115c87e48f43ce9e830d2684b1b9a77c207a75a326a",
|
| 28 |
+
"run_id": "cosmos3_nano_stage_one_lora_r256_replicate_only_500k_step_008000_20261105",
|
| 29 |
+
"schema": "cosmos_trainable_overlay_backup_context_v1",
|
| 30 |
+
"source_commit": "ac7886e496552d8d50573ec3d9157be337ceb7cf",
|
| 31 |
+
"source_run": "/scratch/shared/engs2907/logs/vgg_forecast_4d/train/cosmos3_nano_five_stream_stage_one_lora_r256_replicate_only_500k_20261103/attempt04",
|
| 32 |
+
"source_windows_config": "source_windows.yaml",
|
| 33 |
+
"source_windows_sha256": "ae49ec6d723283ac0429363143cc6e7cebba2e79bdaf8a38c7540f880531dffd",
|
| 34 |
+
"tensor_count": 530,
|
| 35 |
+
"training_log": "/scratch/shared/engs2907/logs/vgg_forecast_4d/slurm/train/cosmos3_nano_five_stream_stage_one_lora_r256_replicate_only_500k_20261103/attempt04/slurm-72961-2.out",
|
| 36 |
+
"training_parameters": {
|
| 37 |
+
"global_batch": 32,
|
| 38 |
+
"gradient_accumulation": 1,
|
| 39 |
+
"lora_alpha": 256,
|
| 40 |
+
"lora_rank": 256,
|
| 41 |
+
"per_rank_batch": 16,
|
| 42 |
+
"streams": [
|
| 43 |
+
"video",
|
| 44 |
+
"pointmap",
|
| 45 |
+
"camera",
|
| 46 |
+
"dino",
|
| 47 |
+
"track_video"
|
| 48 |
+
],
|
| 49 |
+
"training_frames": 9,
|
| 50 |
+
"world_size": 2
|
| 51 |
+
},
|
| 52 |
+
"training_step": "72961.2",
|
| 53 |
+
"training_update": 8000
|
| 54 |
+
}
|