YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

kotodama-3b-trajectory

Training trajectory of the kotodama 3B base (Block AttnRes DD-3B boundaries 0,1,3,7,15,19,24; Muon + AdamW; NCA pre-pretraining then 384B language tokens). Final model: aethera-gp/kotodama-3b-base-final.

  • trajectory/: language-phase checkpoints (steps 32174 .. 190200), torch.save dicts, zstd where .zst
  • milestones/: NCA phase-3 base (the language init), step 22400, backup step 32000, chinchilla-decay branch base (step 30600)
  • metrics.jsonl / geo_metrics.jsonl: training + geometric-monitor logs

Load with kotodama.ckpt.load (github.com/LuxiaSL/kotodama). Note: RMSNorm/QK-norm gains in this run are bit-frozen at 1.0 (pure-bf16 + stock AdamW); fixed for later runs.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support