In this file we list each member's contribution for the project.
Gonzalez, Sergio:
- Initial cluster setup (relocating conda installation to overcome disk space limitations, open ports)
- Defined methodology for Multi-task learning (Loss function, dataloader)
- Updated evaluation.py to standardize inputs as single bert input_ids instead of pairs
- Modified datasets.py and bert.py from part 1 to concatenate sentences and create updated attention_masks with tokenizer
- Experimented with annealed sampling
- Contributed with assigning task
Ksiezak, Kacper:
- Contributed to bert.py encoding part, attention and add_norm
- contributed to forward method of multitask_classifier
- experiments with scheduler
- contributed to conducting the code according to the guidelines
- contributed to training function of multitask classifier
- general debugging of code
- contributed to methodology part
- performing sanity checks of the model
Aldahi, Hasan:
- Contributed to general debugging, fixing coding errors, managing git file versions. *Added and enhanced some of the code functionalities. *Contributed to the Analysis of best model by adding Meta Data for each model. *Added Predictmodel.py for prediction and formatted the output files in the evaluation.py file. *Added the Metrics for having the best model. *Experimenting multiple coding and methodologies and approaches to implement. *Accessing the Cluster and make the word CUDA functional. *Contributed in the checking the architecture of the transformer paper and its implementation.
- Debugging and Adding the necessary codes to implement Bert.py. *Enhanced the code readability by adding comments and deleting redundancy
Tran Ortega, Daniel:
- Wrote README file
- Contributed to setting up computer cluster and running jobs on it
- Contributed to general debugging and fixing coding errors
- Contributed to write forward & prediction methods (multitask_classifier.py) as well as forward, attention and add_norm (bert.py) methods
- Created AI-Usage card
- Contributed to evaluation of models and experimenting approaches
- Contributed to Experiment section
Kemal, Alp Sezer:
- creating optimizer