Releases: tum-ai/number-token-loss
Releases · tum-ai/number-token-loss
ICML 2025
What's Changed
- setup: expand .gitignore to include more Python-related files by @SeanSdahl in #21
- New implementation of wasserstein loss based on the absolute difference by @NinaWie in #22
- Code packaging by @jannisborn in #23
- adjusted paths for new packaging by @zausin33 in #25
- Poster by @jannisborn in #27
- Bump torch from 2.0.1 to 2.2.0 by @dependabot[bot] in #29
- Bump aiohttp from 3.10.5 to 3.10.11 by @dependabot[bot] in #30
- chore: add 'outputs/' to .gitignore by @SeanSdahl in #31
- test: update language modelling tests to enable running on apple silicon by @SeanSdahl in #32
- added code to generate arithmetic expression datasets by @SeanSdahl in #28
- Bump jinja2 from 3.1.4 to 3.1.5 by @dependabot[bot] in #33
- Benchmarking Suite for Loss Functions in T5 Training by @s1k0ra in #36
- Gce by @zausin33 in #39
- Ablation studies by @AnamarijaKozina in #37
- Non-number token loss contributes now to GCE by @ad045 in #41
- Ablation studies without data by @zausin33 in #50
- Integration of base ntl and minimal working example by @Larspennig in #53
- Add Lightweight Runtime Benchmarking for Loss Functions by @s1k0ra in #54
- Multigpu compatibility (#1) by @jannisborn in #57
- Gh page by @zausin33 in #59
- Rebuttal by @zausin33 in #62
- downloaded the cnn dataset and preprocess by @zausin33 in #60
- Profiling by @zausin33 in #61
- added generation file for temperature dataset by @zausin33 in #58
- Gaussian smoothing with NTL by @Larspennig in #64
- Gh page by @zausin33 in #63
New Contributors
- @SeanSdahl made their first contribution in #21
- @dependabot[bot] made their first contribution in #29
- @s1k0ra made their first contribution in #36
- @ad045 made their first contribution in #41
Full Changelog: neurips_mathai...icml25
Contributors
jannisborn, s1k0ra, and 7 other contributors
Assets 2
neurips_mathai
Repo state after publication at NeurIPS MathAI 2024 workshop: https://openreview.net/forum?id=Cb8RP9KLyh
What's Changed
- pipeline first version by @zausin33 in #1
- Add option to train BPE tokenizer from scratch in xVal tokenizer by @annaKett in #2
- Merging T5 Backbone branch into main by @SAint7579 in #7
- added training by @Larspennig in #8
- Xval pipeline by @zausin33 in #9
- Number evaluation by @zausin33 in #10
- Kacper by @Coldyz in #11
- Final eval temp vinc by @zausin33 in #14
- Finalizing eval by @zausin33 in #13
- Data augmentation by @jannisborn in #15
- Finalizing eval by @zausin33 in #17
- Plots by @jannisborn in #16
- added multiplication dataset by @zausin33 in #18
- Cleanup by @zausin33 in #19
- removed unused files by @zausin33 in #20
New Contributors
- @zausin33 made their first contribution in #1
- @annaKett made their first contribution in #2
- @SAint7579 made their first contribution in #7
- @Larspennig made their first contribution in #8
- @Coldyz made their first contribution in #11
- @jannisborn made their first contribution in #15
Full Changelog: https://github.com/tum-ai/number-token-loss/commits/neurips_mathai
Contributors
jannisborn, SAint7579, and 4 other contributors