8bis-lane-models

Part of the 8b.is AI training family: weights, corpora and evaluations produced on hardware we own, kept reproducible and local-first. Nothing here is rented inference, and no claim below is a benchmark unless it says so.

repository	PeetPedro/8bis-lane-models
contents	metadata only
files	4

Family · 8b.is · classroom · the library · github

8bis-lane-models — the constellation's finished voices

Every trained pick that left the oven, with its GitHub source and its seat. A model is a model everywhere — the calibration IS the corpus.

seat	what it is	github source
sim-model	the corridor's deterministic fixture (seed 41592, 6 tensors)	8b-is/training-pipeline
sprint-fixtures (in the corpus repo)	runs/world/mem8-diet parquet + jsonl	8b-is/training-pipeline
upcoming	each ✓ pick seats as <base>-lane-e<effort>-s<steps>-<ts> with its card + run.json	8b-is/8b-is-engine (the engine)

The naming is the bucket convention: save_to_hf.sh seats <base>-lane-<tag>; every card carries both the GitHub source and the HF seat so the two stay linked. the constellation · 0 + 1 · fine touch from within · vaked.dev