BenchFormers v0.4.0

Features

7a7644b feat: Implemented the code for fitting and plotting the power law

Bug Fixes

1cfafb4 fix: Fixed the local model not being loaded
13fa50d fix: Pad only depending on the largest sequence

Documentation

Refactors

Chores

Other Changes

BenchFormers v0.3.1

Features

e2efd73 feat: Added ModernBERT to the list of Models

Bug Fixes

967dcd2 fix: Limited modernbert tokenizer to 4096 length
c8714f9 fix: Get the correct model for modernbert choice
9fc4c88 fix: Passed context window size
dce4ef0 fix: Check if checkpoints array is not empty

Documentation

Refactors

8c871a0 refactor: Debug log statement
27589c6 refactor: Save every 300 steps

Chores

Other Changes

BenchFormers v0.3.0

Features

d2b2cc2 feat: Trainer will look for a checkpoint in the results directory
ace34d9 feat: Re-wrote the code to follow the original split provided by the dataset

Bug Fixes

d0d1ec7 fix: Skip label encoding as it's done before model training

Documentation

Refactors

292adfa refactor: Removed non-alpha-numeric characters from the texts

Chores

Other Changes

BenchFormers v0.2.3

Features

Bug Fixes

59ccc16 fix: Removed unaveraged eval scores to avoid an error

Documentation

Refactors

Chores

Other Changes

BenchFormers v0.2.2

Features

903152e feat: validation code is now using Macro F1-Score instead of weighted one

Bug Fixes

Documentation

Refactors

cb02594 refactor: Also display f1 score for the sake of comparison
4c0e9d1 refactor: Renamed a unit test to match te module it's testing

Chores

Other Changes

BenchFormers v0.2.1

Features

3903845 feat: Seperate a set with 25.000 rows for final validation for better model evaluation

Bug Fixes

Documentation

Refactors

Chores

e800315 chore: Removed unaccessed torch import

Other Changes

d0c5eaa git: Ignore txt files

BenchFormers v0.2.0

Features

e7ea6a8 feat: Made it possible to choose a model to train
fafab10 feat: Set pytorch env variable to allocate less gpu memory

Bug Fixes

d4b8aa8 fix: Used --models argument
a6f7563 fix: Return the model as a list to make it iterable
a6e72ac fix: Removed the unused loop
8f5d07d fix: Typo
c9c3036 fix: Correct usage of main.py
26eff72 fix: Brought back Roformer Model
7ee941f fix: Added gradient checkpointing to save GPU memory
e942843 fix: syntax error on logger
63febf1 fix: Removed the problematic line of code
3c969f3 fix: Unterminated string

Documentation

Refactors

c36f594 refactor: Models will be trained sequentially to save gpu memory usage
8e35ee7 refactor: Optimised trainer to use less GPU memory
e8231c9 refactor: Stratify the labels
a05e669 refactor: Formatted code as per PEP requirements
71ce5c2 refactor: Removed unnecessary parameters

Chores

0e654d6 chore: Added targets for evaluation
01e5f0e chore: update logger to default to info
310cd4f chore: Added target for 1k data
a3f9dad chore: Added targets for running the code

Other Changes

cbc3701 git: Workflow that drafts a release when tag is pushed