Commits

Branch selector

 master

User selector

All users

All time

Commit History

Commits on Oct 13, 2025

• Merge pull request #32 from antoniorv6/eric-ayllon-patch-1

  Show description for aad2b20
  eric-ayllon

  authored
  aad2b20

  Copy full SHA for aad2b20

• Update configuration_smt.py

  Show description for 310e15a
  eric-ayllon

  authored
  310e15a

  Copy full SHA for 310e15a

Commits on Jul 23, 2025

• Silly update of README.md

  Show description for 6d66123
  antoniorv6

  authored
  6d66123

  Copy full SHA for 6d66123

Commits on Jul 18, 2025

• Merge pull request #27 from antoniorv6/bugfix

  Show description for 5ea08ea
  antoniorv6

  authored
  5ea08ea

  Copy full SHA for 5ea08ea

• Enabled multiple workers for the training dataloader.

  eric-ayllon

  committed
  961c517

  Copy full SHA for 961c517

Commits on Jul 7, 2025

• Removed a redundant loop and simplified the kern parsing for calculating the poliphony metrics.

  eric-ayllon

  committed
  8023c8c

  Copy full SHA for 8023c8c

Commits on Jul 4, 2025

• Pending to test whether the data loading bug is fixed and did not break something else.

  eric-ayllon

  committed
  e921ec4

  Copy full SHA for e921ec4

• Added checkpointer configs to config files of other datasets to prevent the ExperimentConfig from raising an error.

  eric-ayllon

  committed
  ee28d88

  Copy full SHA for ee28d88

Commits on Jun 19, 2025

• Fixed README.md tip bug

  antoniorv6

  authored
  1ab56dc

  Copy full SHA for 1ab56dc

• Updated README.md links and presentation

  antoniorv6

  authored
  4514276

  Copy full SHA for 4514276

• Merge pull request #21 from antoniorv6/docs-update

  antoniorv6

  authored
  8a799c6

  Copy full SHA for 8a799c6

• Updated README.md with new references

  antoniorv6

  authored
  4b5ef8f

  Copy full SHA for 4b5ef8f

Commits on Jun 16, 2025

• Merge pull request #20 from antoniorv6/full-page

  Show description for f4a8533
  antoniorv6

  authored
  f4a8533

  Copy full SHA for f4a8533

• Removed logging done from the dataset instances to (probably) prevent wandb from crashing when using workers and the synthetic generator.

  eric-ayllon

  committed
  7b2c530

  Copy full SHA for 7b2c530

Commits on Jun 14, 2025

• Added instructions for running full-page scripts for results reproduction purposes.

  eric-ayllon

  committed
  48446bc

  Copy full SHA for 48446bc

• Restored the stage_skip value to stop skipping curriculum learning stages by default. To skip curriculum learning steps (for example to continue a stopped training) modify the skip_cl_steps variabl…

  Show description for 5ace052
  eric-ayllon

  committed
  5ace052

  Copy full SHA for 5ace052

• Added requirements for full-page experiments.

  eric-ayllon

  committed
  d3408d7

  Copy full SHA for d3408d7

• Removed the hardcoded maximum from the datamodule that only uses data from huggingface.

  eric-ayllon

  committed
  4e8c18d

  Copy full SHA for 4e8c18d

• Changed the system-level training script to retrieve the max height, width and seqlen from the datamodule instead of the datamodule's train set.

  eric-ayllon

  committed
  65c632b

  Copy full SHA for 65c632b

• Moved the max height, width and seqlen calculations to the datamodule.

  eric-ayllon

  committed
  4303cf5

  Copy full SHA for 4303cf5

Commits on Jun 13, 2025

• Merge branch 'master' into full-page

  eric-ayllon

  authored
  fadbfc8

  Copy full SHA for fadbfc8

Commits on Jun 12, 2025

• Changed the compute_regular_attention so that the attention weights are returned after applying the softmax rather than before. This change was made to keep backwards compatibility.

  eric-ayllon

  committed
  a773c7d

  Copy full SHA for a773c7d

• Fixed return_weights and compute_regular_attention.

  eric-ayllon

  committed
  f83b84b

  Copy full SHA for f83b84b

• Forced images to be converted to RGB to keep consistency in data augmentation.

  eric-ayllon

  committed
  8fa0352

  Copy full SHA for 8fa0352

Commits on Jun 10, 2025

• Returned to 0 workers on the trainer because the synthetic generator is quite problematic when used with multithreading.

  eric-ayllon

  committed
  e02b46e

  Copy full SHA for e02b46e

Commits on Jun 9, 2025

• Added the experiments for Mozarteum and Polish_Scores.

  eric-ayllon

  committed
  f18796d

  Copy full SHA for f18796d

Commits on Jun 6, 2025

• Renamed the config folders of full-page experiments to have FP- in the name.

  eric-ayllon

  committed
  a9a39d7

  Copy full SHA for a9a39d7

• Added Mozarteum and Polish_Scores experiment configs.

  eric-ayllon

  committed
  ce94f86

  Copy full SHA for ce94f86

• I forgot to restore the workers after debugging.

  eric-ayllon

  committed
  ba0dddf

  Copy full SHA for ba0dddf

Commits on Jun 5, 2025

• Added the option to skip curriculum steps for when training takes too long and has to be split in multiple executions.

  eric-ayllon

  committed
  76a3419

  Copy full SHA for 76a3419

Commits on Jun 4, 2025

• Merge pull request #17 from antoniorv6/fix-data

  antoniorv6

  authored
  a824694

  Copy full SHA for a824694

• fix(data): Fix data resizing no application

  antoniorv6

  committed
  b2b800c

  Copy full SHA for b2b800c

Commits on Jun 2, 2025

• Fixed some data loading errors and also added a constraint to force the trainer to make at least 300k steps to complete the curriculum learning stage and reduce the synthetic sample probability dow…

  Show description for 2371404
  eric-ayllon

  committed
  2371404

  Copy full SHA for 2371404

Commits on May 30, 2025

• Fixed some things but the synthetic generator throws some rythm analysis errors.

  eric-ayllon

  committed
  bb9783f

  Copy full SHA for bb9783f

• I had forgotten to assign the result of the map.

  eric-ayllon

  committed
  3c8bedd

  Copy full SHA for 3c8bedd

Pagination

Previous

Next