Add expansion metadata
Browse files
expanded-from-scugnizz-1b/expand_metadata.json
ADDED
|
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"created_at": "2026-07-04 08:22:51",
|
| 3 |
+
"target_params": 2648194560,
|
| 4 |
+
"source_repo_id": "Daisuke675/scugnizz-1b",
|
| 5 |
+
"source_ckpt": "training-runs/pretrain-1b/checkpoint_best.pt",
|
| 6 |
+
"target_repo_id": "Daisuke675/scugnizz-3b",
|
| 7 |
+
"target_path": "expanded-from-scugnizz-1b/checkpoint_init_3b.pt",
|
| 8 |
+
"old_config": {
|
| 9 |
+
"vocab_size": 50257,
|
| 10 |
+
"block_size": 1024,
|
| 11 |
+
"n_layer": 24,
|
| 12 |
+
"n_head": 16,
|
| 13 |
+
"n_embd": 2048,
|
| 14 |
+
"dropout": 0.1,
|
| 15 |
+
"bias": false,
|
| 16 |
+
"pcs_a": 0.8309193524478643,
|
| 17 |
+
"pcs_b": 0.0
|
| 18 |
+
},
|
| 19 |
+
"new_config": {
|
| 20 |
+
"vocab_size": 50257,
|
| 21 |
+
"block_size": 1024,
|
| 22 |
+
"n_layer": 32,
|
| 23 |
+
"n_head": 20,
|
| 24 |
+
"n_embd": 2560,
|
| 25 |
+
"dropout": 0.0,
|
| 26 |
+
"bias": false,
|
| 27 |
+
"pcs_a": 0.8309193524478643,
|
| 28 |
+
"pcs_b": 0.0
|
| 29 |
+
},
|
| 30 |
+
"warning": "Expanded checkpoint requires continued pretraining; do not treat as trained 3B model."
|
| 31 |
+
}
|