Spaces:
Sleeping
Sleeping
Commit History
-
update model to 0612
5d45649 -
Merge branch 'main' into add-hf-space
f92296b -
Merge pull request #24 from HEP-KBFI/jp_20260612_dl_benchmark
Laurits TaniUnverified signatureDataloader optimizationsd335060 -
update test
e678f44 -
Merge branch 'main' into jp_20260612_dl_benchmark
Unverified signaturea28f1b6 -
Merge pull request #23 from HEP-KBFI/jp_20260611_scaling
Unverified signatureScaling studyd5c27d0
-
large stats
Joosep Patab8483f6
-
dataset
Joosep Pata1cf17d6 -
io, training: interleave row groups and fix multi-task loss aggregation
- Implemented signal/background row-group interleaving in ParT_dataloader to ensure every batch has a representative sample of all tasks. - Enabled shuffling for validation/test sets for better BatchNorm stability. - Fixed multi-task loss logging in MultiParTau_module by correctly weighting epoch-level averages by the number of contributing jets. - Cleaned up manual loss accumulation logic in favor of self.log.a805ee9 -
large stats training
Joosep Pata32b92f7 -
io: implement intra-chunk shuffling and optimize scaling in ParT_dataloader
- Added intra-chunk shuffling to ParT_dataloader to fix poor label mixing between signal and background row groups, resolving performance discrepancies compared to preprocessed dataloaders. - Optimized scaling logic to pre-load the scaler once instead of per-batch. - Added automatic scaler fitting in ParTDataModule.setup if missing.68b9a50 -
Add row_groups_per_read to dataloader config to allow Hydra overrides
d02f338 -
make dataloader configurable
ad06b5c -
update ml-tau-data version
8660e62 -
update ml-tau-data with p4 optimization
b56dc63 -
Optimize raw dataloader with batched Parquet reading and metadata caching
966f70c -
Add dataloader unit tests and benchmarks
4b26ad2 -
Merge pull request #22 from HEP-KBFI/jp_20260611_dataloader
Laurits TaniUnverified signatureUnify preprocessing to avoid inconsistenciesf627b91 -
fix description
4ab00d4 -
update docs
a433f25 -
scaling study extraction
8b3451e
-
scaling study
c11be62 -
scaling study
392852c -
improve visualization
0ec8f60 -
docs fixes
Joosep Pata12ed270 -
fix
f2e661e -
Merge branch 'jp_20260611_dataloader' into add-hf-space
717adea -
removed app
c3b97bf -
fix ml-tau-data
4427124 -
Merge branch 'jp_20260611_dataloader' into add-hf-space
75a4467 -
unify preprocessing from ml-tau-data
e82ed36 -
Merge branch 'main' into add-hf-space
c3ce6d0 -
Merge pull request #13 from HEP-KBFI/jp_20260609_simpligy
Laurits TaniUnverified signatureSimplify MultiParTau1a17541 -
remove phase2
9c0f3ed -
Merge branch 'jp_20260609_simpligy' into add-hf-space
a9def6c -
revert run script
b24c2e1 -
revert submit script
cbce44a -
added model badge
16c807a -
updated model to 0609_multipartau_full_fef7149
a79ee0c -
Merge branch 'jp_20260609_simpligy' into add-hf-space
8c9afa5 -
update docs
dd3202b -
update model hash
Joosep Pata30ee4f4
-
task weight scheduler
fef7149 -
confusion matrix labels
bf25495 -
update visualization
5cb3880 -
update visualization
c4b4537 -
Merge branch 'jp_20260609_simpligy' into add-hf-space
4e210c0 -
added HF upload script
f327745
-
only multi
Joosep Pataaf76c4c -
share cls_blocks
bc477b9