data/train_init_weights.json

{
  "version": 5,
  "name": "teutonic-II-110B-finewebedu-only",
  "description": "TRAIN mix for Teutonic-II-110B: FineWeb-Edu only (MiMo-V2.5-Pro tokenizer, seq_len=2048). Source of truth is the live R2 datasets catalog.",
  "formula": "Single source finewebedu weight=1.0 matching official competition catalog.",
  "datasets_manifest_url": "https://pub-fedac496355c4edc9aed57189e6e190f.r2.dev/datasets/manifest.json",
  "eval_anchor": {
    "note": "Official Teutonic-II-110B catalog (generated_at 2026-08-24). FineWeb-Edu proportion=1.0, eval_n=2000, delta_threshold=0.5.",
    "finewebedu": 1.0
  },
  "name_map_to_local": {
    "finewebedu": "finewebedu"
  },
  "weights": {
    "finewebedu": 1.0
  }
}