bussin-400m ·
396,418,176 parameters · updated 2026-10-19T11:24:39.855065Z
· page refreshes every 3 min
2.2 tokens/parameter — below minimally trained. Output will be incoherent.
vs Chinchilla (20 tok/param): 11% · ETA 4.9 weeks · weekly capacity 5.9B tokens
This is the panel that says whether the run is on track. Loss alone does not.
| platform | used | remaining | resets in | available | mode | |
|---|---|---|---|---|---|---|
| kaggle-tpu | 0.0 / 20 h | 18.0 h | 109 h | yes | auto | |
| kaggle-gpu | 20.0 / 30 h | 7.0 h | 109 h | yes | auto | |
| lightning | 0.0 / 15 h | 13.5 h | 301 h | yes | auto | |
| colab | 0.0 / 12 h | 10.8 h | 13 h | yes | auto |
Quota is estimated: Kaggle does not expose it via API, so the orchestrator accrues it from dispatch durations against a 90% safety margin. Reconcile against the settings page occasionally.
| worker | platform | steps | end step | efficiency |
|---|---|---|---|---|
| none yet | ||||
| ETL stage | status |
|---|---|
| mine-0 | COMPLETE |
| mine-1 | COMPLETE |
| mine-2 | COMPLETE |
| mine-genz-0 | COMPLETE |
| mine-genz-1 | COMPLETE |
| mine-genz-2 | COMPLETE |
| mine-genz-3 | COMPLETE |
| mine-genz-4 | COMPLETE |
| rank | CANCEL_ACKNOWLEDGED |
| rank-fast | COMPLETE |
| shard | CANCEL_ACKNOWLEDGED |
| shard-0 | COMPLETE |
| shard-1 | COMPLETE |
| shard-2 | COMPLETE |
| shard-3 | COMPLETE |
| shard-4 | COMPLETE |
| shard-val | COMPLETE |
| tokenizer | COMPLETE |
corpus GodAsap/bussin-corpus · 16 shards · 12.36 GB · updated 2026-10-18T07:43:33Z
ETL runs on Kaggle CPU sessions, which consume no GPU quota and so never appear in the ledger above.
No results yet. Pretraining reports loss and perplexity, not accuracy — accuracy exists only once BUSSBENCH runs, which costs GPU time and so happens at milestones.