Upload train_manifest.json with huggingface_hub
Browse files- train_manifest.json +18 -0
train_manifest.json
ADDED
|
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"rows": 764626,
|
| 3 |
+
"truncation": 0,
|
| 4 |
+
"trunc_pct": 0.0,
|
| 5 |
+
"targets": {
|
| 6 |
+
"v3_target": 653386,
|
| 7 |
+
"v4_target": 111240,
|
| 8 |
+
"shared": 647799
|
| 9 |
+
},
|
| 10 |
+
"v4_degenerate_dropped": 0,
|
| 11 |
+
"mix": {
|
| 12 |
+
"strategy": "per_domain",
|
| 13 |
+
"decided_by": "gold_table.json (v3 probe v3, 2026-10-10)",
|
| 14 |
+
"will_instruction": "per domain, highest-quality source; one epoch; frequent checkpoints"
|
| 15 |
+
},
|
| 16 |
+
"yuri_v1_note": "kept for training (valid rows); eval gates exclude D1 per their convention",
|
| 17 |
+
"tokenizer": "/home/will/uncensored-jev/merged/huihui-lm"
|
| 18 |
+
}
|