build1 distinct publisher
The weighting ran, then the dedup pass undid it 56 lines later
MLX LoRA has no per-example weight field, so one builder encoded his curriculum as duplicate lines. A dedup key on the last 200 characters deleted 38,988 of them before training.
Publishers:dev.to
Reality
- Evidence63
- Adoption14
- Hype gap−9
- Incentives32
- Confidence57