Skip to content

Easy to maintain, modular version of 🤗run_glue.py

Notifications You must be signed in to change notification settings

kainoj/run_glue

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

40 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Easy to maintain, modular version of 🤗run_glue.py

👉 See wandb.ai dashboard here 👈


Evaluate your custom model in 3 steps:

  1. Put code of your model under src/models/
  2. Configure it under configs/model/your_model.yaml
  3. python run_glue.py model=your_model and open your wandb dashboard.

Evaluate in bulk (no need for bash looping)

python run_glue.py --multirun \
    experiment=cola,wnli,mrpc \
    model.learning_rate=2e-5,3e-5

Will run 3*2 = 6.

Make GPUs do vroom-voom 🏎

python run_glue.py --multirun \
    experiment=cola,wnli,mrpc \
    trainer.gpus=2 \           # 2 gpus per task
    trainer.precision=16 \     # Mixed 16FP precision
    datamodule.batch_size=256  # Beefy batch size

Setup

conda env create -f environment.yml
conda activate run-glue

Credits

  • Lightning-hydra template by ashleve.
  • Some functionalities taken from the original run_glue.py.

About

Easy to maintain, modular version of 🤗run_glue.py

Resources

Stars

Watchers

Forks

Languages