Skip to content

akuritsyn/feedback-prize-2021

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

11 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Top 2% solution (36/2060) based on a baseline model developed by Abhishek Thakur.

Requirements

Architecture

  • The solution is an ensemble of 5 transformer models, each trained on 5 folds: 2x deberta-large, 2x deberta-v3-large and 1x longformer-large. The two versions of deberta models were trained using dropout=0.1 and dropout=0.15 and max_len=1024 parameters, while for longformer-large model dropout=0.1 and max_len=1536 parameters were used.

Preparation

  • Download tez pytorch trainier library and put it at the root level
  • Put ./data directory at the root level and unzip the files downloaded from Kaggle there.
  • In order to use deberta v2 or v3, you need to patch transformers library to create a new fast tokenizer using data and instructions from this kaggle dataset.
  • Download microsoft/deberta-large, microsoft/deberta-v3-large and allenai/transformer-large or any other transformer models using nbs/download_model.ipynb and save them in ./model folder.
  • Create 5 training folds using nbs/creating_folds.ipynb.

Training

Please make sure you run the script from parent directory of ./bin.

$ sh ./bin/train.sh

To train different models on different folds (0...4) make changes inside the train.sh file.

The training of each fold should fit into 15GB GPU memory.

Inference

About

Kaggle NLP competition - Top 2% solution (36/2060)

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published