Multi-Task Learning in codemix environment

Project Structure

.
├── base_models
├── data
│    └── lince
│       └── ner
|    └── IPM
├── runs
├── notebooks
├── src
│    └── datamodules
│    └── models
│          └── baseline
│          └── metalearning
│    └── modules
│    └── utils
├── .gitignore
├── config.py
├── main.py 
└── README.md

Configure Setup

Use conda env create -f env.yml command to first create the environment.

How to run ?

# Hyperparams
parser.add_argument("--epochs", type=int, default=MAX_EPOCHS, help="Set max epochs")
parser.add_argument("--lr", type=float, default=LEARNING_RATE, help="Set Learning Rate")
parser.add_argument("--weight_decay", type=float, default=WEIGHT_DECAY, help="Set Weight Decay")
parser.add_argument("--dropout", type=float, default=DROPOUT_RATE, help="Set dropout rate")
parser.add_argument("--max_seq_len", type=int, default=MAX_SEQUENCE_LENGTH, help="Set max seq length")
parser.add_argument("--padding", type=str, default=PADDING, help="Set padding style")
parser.add_argument("--batch_size", type=int, default=BATCH_SIZE, help="Set batch size")
parser.add_argument("--base_model", type=str, default=BASE_MODEL, help="Set base transformer model")
parser.add_argument("--unfreeze", type=bool, default=None, help="Freeze or Unfreeze base model")
parser.add_argument("--dataset", type=str, default="lince", help="Set dataset to be used")

# Hardware
parser.add_argument("--workers", type=int, default=NUM_WORKERS, help="Set CPU Threads")
parser.add_argument("--gpus", type=int, default=AVAIL_GPUS, help="Set no. of GPUs required")