📦 WP-079 — fix(data): per-worker, per-epoch augmentation RNG¶
Research log · Phase 6 — Evaluation, v0.1.0
Engineering log · Phase 6 — Evaluation, v0.1.0
WP-079 — per-worker, per-epoch augmentation RNG: what the collision cost¶
Measured exposure: 1/3200 of the intended draws at 32 workers over 100 epochs. Two DataLoader defects compose multiplicatively and neither is visible in a loss curve — the run trains, it simply trains on far less augmentation variety than the config describes. The mechanism is a _TrainPipeline bug fixed the same day; the full story is at wp-079.md#wp-079-engineering.
Engineering log · Phase 6 — Evaluation, v0.1.0
WP-079 — per-worker, per-epoch augmentation RNG¶
_TrainPipeline seeded one generator in the parent process. A parent's generator never advances when workers are used, so every worker replayed one identical augmentation-parameter stream — diversity 1/num_workers. WP-076 then made workers non-persistent, so the pool restarted from that same state every epoch, replaying a single epoch's parameters for the entire run.
What the two defects together cost the training-data variety is measured at wp-079.md#wp-079.