Add immutable checkpoint backup cosmos3_nano_full20_21frame_update_010000_20261106 run_execution_metadata
Browse files
cosmos3_nano_stage_one_lora_r256_full20_21frame_ddp_update_010000_20261106/attempt01/update_010000/backup_context.json
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": "cosmos3_nano_trainable_overlay_backup_context_v1",
|
| 3 |
+
"run_id": "cosmos3_nano_full20_21frame_update_010000_20261106",
|
| 4 |
+
"repository": "yslan/o-intermediate-model",
|
| 5 |
+
"remote_prefix": "cosmos3_nano_stage_one_lora_r256_full20_21frame_ddp_update_010000_20261106/attempt01/update_010000",
|
| 6 |
+
"checkpoint_original": "/scratch/shared/engs2907/logs/vgg_forecast_4d/eval/cosmos_midtraining_update10000_20261106/attempt01/step_010000.safetensors",
|
| 7 |
+
"checkpoint_sha256": "e1c67cc013c2cc3acc52fbcaed098b3dfe89cc4adb2d7c08944dfb5ab7cd68b1",
|
| 8 |
+
"checkpoint_bytes": 2990807176,
|
| 9 |
+
"checkpoint_kind": "trainable LoRA and projector parameters only",
|
| 10 |
+
"optimizer_step": 10000,
|
| 11 |
+
"base_model": "nvidia/Cosmos3-Nano@e59a53c25979a090fa8706c9acc0c254a6e89b92",
|
| 12 |
+
"cosmos_training_source_commit": "8e8331d4e101005eb362e2fea713a19bd32c355a",
|
| 13 |
+
"vgg_source_pin_commit": "b9e2059143b4db90a57a4bac7d5dca1a7ee03d48",
|
| 14 |
+
"source_windows_sha256": "0b35b990c82717b4d998691e616581573679ca53f30c44dc370577362b31905f",
|
| 15 |
+
"config_identity": "cosmos3_nano_omni4d_stage_one_lora_r256_plain_ddp_21_frames_temporal_forcing_full20_from_overlay_global32_lr2p5e5",
|
| 16 |
+
"warm_start": {"update": 7700, "sha256": "b46067d50dc854c225c1570da2c25691bfad1d98dd97436af6f2c5cf1ec02b9b", "kind": "weights-only with fresh optimizer", "parent_checkpoint_entry": "cosmos3_nano_full19_21frame_update_007320_hf_backup_20261105"},
|
| 17 |
+
"training": {"source_count": 20, "frames": 21, "world_size": 2, "per_rank_batch": 16, "gradient_accumulation": 1, "global_batch": 32, "learning_rate": 0.000025, "precision": "bf16", "lora_rank": 256, "max_updates": 500000, "temporal_forcing": true},
|
| 18 |
+
"eval_lineage_file": "evaluation_lineage.json",
|
| 19 |
+
"excluded": ["frozen pretrained base weights", "optimizer state", "RNG state", "full model weights"],
|
| 20 |
+
"preserve_existing_remote_paths": ["stage1_4d_wan22_5b_step_015000/", "all existing 7.3K and 15K backup paths"]
|
| 21 |
+
}
|