May 22 21:48 length-bucketed training and HF bootstrap 02:17 Fix AMP control in training loop 00:49 allow disabling amp for hf training May 21 23:49 skip invalid simulator loci 18:04 remove local metadata artifact 18:03 sync current ml-intern source for hf jobs May 19 19:09 Harden missing-label handling in splits and dataset loading May 16 02:05 Preserve completed runs from overwrite May 15 18:38 Reduce validation GPU memory and reuse LMDB handles 15:47 Fix LMDB reuse during validation May 14 19:37 Support pre-split parquet training inputs Allow training ETL to consume train/val/test parquet roots directly so remote jobs can reuse existing split datasets instead of rebuilding the split map from raw parquet. 19:27 Reduce parquet split-map memory Build the family split map from the label column only so remote HF Jobs do not materialize the full sequence column before ETL. 19:25 Sync DeepGenopix package snapshot Upload the current src/deepgenopix package so remote HF Jobs use a consistent code snapshot, including the analysis/notebook_support pair and proportional parquet family holdouts. 19:07 Use proportional family holdouts for parquet ETL Make train/val/test holdouts proportional per TE family so validation and test splits are not over-allocated for mid-sized families. 18:22 Update src/deepgenopix/notebook_support.py from ml-intern branch 18:22 Update src/deepgenopix/cli.py from ml-intern branch May 4 21:12 Upload src/deepgenopix/etl_streaming.py 20:37 Upload src/deepgenopix/cli.py 20:37 Upload src/deepgenopix/__init__.py 02:53 Upload src/deepgenopix/quant.py 02:31 Upload src/deepgenopix/analysis.py 02:29 Upload src/deepgenopix/inference.py 02:28 Upload src/deepgenopix/trainer.py 02:26 Upload src/deepgenopix/model.py 02:24 Upload src/deepgenopix/dataset.py 02:23 Upload src/deepgenopix/etl.py 02:20 Upload src/deepgenopix/io_utils.py 02:19 Upload src/deepgenopix/pixelizer.py 02:12 Upload src/deepgenopix/config.py 02:11 Upload src/deepgenopix/__init__.py 
