Sparse Progressive Distillation: Resolving Overfitting under Pretrain-and-Finetune Paradigm

Last update: Dec 5, 2022

Related tags

Deep Learning SparseProgressiveDistillation

Overview

Sparse Progressive Distillation: Resolving Overfitting under Pretrain-and-Finetune Paradigm

This is the Pytorch implementation for sparse progressive distillation (SPD). For more details about the motivation, techniques and experimental results, refer to our paper here.

Running

Environment Preparation (using python3)
```
pip install -r requirements.txt
```
Dataset Preparation

The original GLUE dataset could be downloaded here.

BERT_base fine-tuning on GLUE

We use finetuned BERT_base as the teacher. For each task of GLUE benchmark, we obtain the finetuned model using the original huggingface transformers code with the following script.

python run_glue.py \
          --model_name_or_path $INT_DIR \
          --task_name $TASK_NAME \
          --do_train \
          --do_eval \
          --data_dir $GLUE_DIR/$TASK_NAME/ \
          --max_seq_length 128 \
          --per_gpu_train_batch_size 32 \
          --per_gpu_eval_batch_size 32 \
          --learning_rate 3e-5 \
          --num_train_epochs 4.0 \
          --output_dir $OUT_DIR \
          --evaluate_during_training \
          --overwrite_output_dir \
          --logging_steps 400 \
          --logging_dir $OUT_DIR \
          --save_steps 10000

Sparse Progressive Distillation

We use run_glue.py to run the sparse progressive distillation. --num_prune_epochs is the epochs for pruning. --num_train_epochs is the total number of epochs (pruning, progressive distillation, finetuning).

python run_glue.py \
  --model_name_or_path PATH_TO_FINETUNED_MODEL \
  --task_name $TASK_NAME \
  --do_train \
  --do_eval \
  --do_lower_case \
  --data_dir $GLUE_DIR/$TASK_NAME/ \
  --max_seq_length 128 \
  --per_gpu_train_batch_size 32 \
  --per_gpu_eval_batch_size 32 \
  --learning_rate 6.4e-4 \
  --save_steps 50 \
  --num_prune_epochs 30 \
  --num_train_epochs 60 \
  --sparsity 0.9 \
  --output_dir $OUT_DIR \
  --evaluate_during_training \
  --replacing_rate 0.8 \
  --overwrite_output_dir \
  --steps_for_replacing 0 \
  --scheduler_type linear

To Dos

Provide our teacher model for each task.
Provide best performed model checkpoint for each task.

Minimal But Practical Image Classifier Pipline Using Pytorch, Finetune on ResNet18, Got 99% Accuracy on Own Small Datasets.

PyTorch Image Classifier Updates As for many users request, I released a new version of standared pytorch immage classification example at here: http:

106 Nov 6, 2022

Sparse Progressive Distillation: Resolving Overfitting under Pretrain-and-Finetune Paradigm

Related tags

Overview

Sparse Progressive Distillation: Resolving Overfitting under Pretrain-and-Finetune Paradigm

Running

BERT_base fine-tuning on GLUE

Sparse Progressive Distillation

To Dos

You might also like...

Minimal But Practical Image Classifier Pipline Using Pytorch, Finetune on ResNet18, Got 99% Accuracy on Own Small Datasets.

Finetune SSL models for MOS prediction

Finetune the base 64 px GLIDE-text2im model from OpenAI on your own image-text dataset

PyTorch implementation of Progressive Growing of GANs for Improved Quality, Stability, and Variation.

Efficient Conformer: Progressive Downsampling and Grouped Attention for Automatic Speech Recognition

Multi-Stage Progressive Image Restoration

Progressive Domain Adaptation for Object Detection

Code for Referring Image Segmentation via Cross-Modal Progressive Comprehension, CVPR2020.

(AAAI 2021) Progressive One-shot Human Parsing

Owner

Re-implementation of 'Grokking: Generalization beyond overfitting on small algorithmic datasets'

PyTorch implementation of Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets

TF2 implementation of knowledge distillation using the "function matching" hypothesis from the paper Knowledge distillation: A good teacher is patient and consistent by Beyer et al.

[IJCAI-2021] A benchmark of data-free knowledge distillation from paper "Contrastive Model Inversion for Data-Free Knowledge Distillation"

Differentiable Neural Computers, Sparse Access Memory and Sparse Differentiable Neural Computers, for Pytorch

This is the official implement of paper "ActionCLIP: A New Paradigm for Action Recognition"

This is the official implement of paper "ActionCLIP: A New Paradigm for Action Recognition"

Supervision Exists Everywhere: A Data Efficient Contrastive Language-Image Pre-training Paradigm

Video Instance Segmentation with a Propose-Reduce Paradigm (ICCV 2021)

ByteTrack with ReID module following the paradigm of FairMOT, tracking strategy is borrowed from FairMOT/JDE.