Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/microsoft/adamix

This is the implementation of the paper AdaMix: Mixture-of-Adaptations for Parameter-efficient Model Tuning (https://arxiv.org/abs/2205.12410).
https://github.com/microsoft/adamix

adapter bert dart glue gpt-2 nlg nlp nlu parameter-efficient pytorch roberta webnlg

Last synced: about 1 month ago
JSON representation

This is the implementation of the paper AdaMix: Mixture-of-Adaptations for Parameter-efficient Model Tuning (https://arxiv.org/abs/2205.12410).

Host: GitHub
URL: https://github.com/microsoft/adamix
Owner: microsoft
License: apache-2.0
Created: 2022-05-24T22:52:02.000Z (over 2 years ago)
Default Branch: main
Last Pushed: 2023-08-14T22:22:39.000Z (about 1 year ago)
Last Synced: 2024-10-07T11:22:33.288Z (about 1 month ago)
Topics: adapter, bert, dart, glue, gpt-2, nlg, nlp, nlu, parameter-efficient, pytorch, roberta, webnlg
Language: Python
Homepage:
Size: 16.7 MB
Stars: 126
Watchers: 5
Forks: 11
Open Issues: 25
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Code of conduct: CODE_OF_CONDUCT.md
- Security: SECURITY.md

Awesome Lists containing this project

README

        # AdaMix (Mixture-of-Adaptations)

This is the implementation of the paper [AdaMix: Mixture-of-Adaptations for Parameter-efficient Model Tuning](https://arxiv.org/abs/2205.12410). 

## Quick links

* [Overview](#overview)

* [GLUE Benchmark](#adapting-to-the-glue-benchmark)

  * [BERT-base](#main-results-bert-base)

  * [RoBERTa-large](#main-results-roberta-large)

  * [Averaging Weigths v.s. Random Routing](#comparision-between-averaging-weights-and-random-single-adapter)

* [Checkpoints](#download-adamix-checkpoints)

* [Run the model](#steps-to-reproduce-our-results)

  * [Quick start](#quick-start)

  * [Evaluate checkpoints](#evaluate-the-checkpoints)

* [Notes and Acknowledgments](#notes-and-acknowledgments)

* [Contact Information](#contact-information)

* [Citation](#how-do-i-cite-adamix)

## Overview

### Training



### Inference



## Adapting to the GLUE Benchmark

Our experiments on the GLUE benchmark are run on 16 NVIDIA Tesla V100 GPU. The results may vary due to different GPU models, drivers, CUDA SDK versions, floating-point precisions, and random seeds. 

### Main Results (BERT-base)



### Main Results (RoBERTa-large)



### Comparision between Averaging Weights and Random Single Adapter 



## Download AdaMix checkpoints

We release all copies of Adapter weights for users' Adapter aggregation study. 

|   | Dataset  | BERT base 110M 
   | RoBERTa large 355M 
  |

|---|----------|--------------------|----------------------|

|   | MNLI     |[8.5 MB](https://github.com/yaqingwang/MoA/releases/download/bert_base/pytorch_model_mnli_expert_soup.bin) |[11.7 MB](https://github.com/yaqingwang/MoA/releases/download/roberta_large/pytorch_model_mnli_expert_soup.bin) |

|   | SST2     |[8.5 MB](https://github.com/yaqingwang/MoA/releases/download/bert_base/pytorch_model_sst2_expert_soup.bin)  |[11.7 MB](https://github.com/yaqingwang/MoA/releases/download/roberta_large/pytorch_model_sst2_expert_soup.bin)  |

|   | MRPC     |[8.5 MB](https://github.com/yaqingwang/MoA/releases/download/bert_base/pytorch_model_mrpc_expert_soup.bin)  |[11.7 MB](https://github.com/yaqingwang/MoA/releases/download/roberta_large/pytorch_model_mrpc_expert_soup.bin)  |

|   | CoLA     |[8.5 MB](https://github.com/yaqingwang/MoA/releases/download/bert_base/pytorch_model_cola_expert_soup.bin)  |[11.7 MB](https://github.com/yaqingwang/MoA/releases/download/roberta_large/pytorch_model_cola_expert_soup.bin)  |

|   | QNLI     |[8.5 MB](https://github.com/yaqingwang/MoA/releases/download/bert_base/pytorch_model_qnli_expert_soup.bin)  |[11.7 MB](https://github.com/yaqingwang/MoA/releases/download/roberta_large/pytorch_model_qnli_expert_soup.bin)  |

|   | QQP      |[8.5 MB](https://github.com/yaqingwang/MoA/releases/download/bert_base/pytorch_model_qqp_expert_soup.bin)  |[11.7 MB](https://github.com/yaqingwang/MoA/releases/download/roberta_large/pytorch_model_qqp_expert_soup.bin)  |

|   | RTE      |[8.5 MB](https://github.com/yaqingwang/MoA/releases/download/bert_base/pytorch_model_rte_expert_soup.bin)  |[11.7 MB](https://github.com/yaqingwang/MoA/releases/download/roberta_large/pytorch_model_rte_expert_soup.bin)  |

|   | STSB     |[8.5 MB](https://github.com/yaqingwang/MoA/releases/download/bert_base/pytorch_model_stsb_expert_soup.bin)  |[11.7 MB](https://github.com/yaqingwang/MoA/releases/download/roberta_large/pytorch_model_stsb_expert_soup.bin)  |

## Steps to reproduce our results

### Create and activate conda env

```console

conda env create -f environment.yml

```

### Install the pre-requisites

```console

pip install -e .

```

We also provide the shell scripts for bert-base and roberta-large.

### Quick start

```console

export num_gpus=1

export PYTHONHASHSEED=0

task_name=mnli

model=roberta-large

export output_dir="./models/${model}/${task_name}"

python -m torch.distributed.launch --nproc_per_node=$num_gpus \

examples/text-classification/run_glue.py \

--model_name_or_path $model \

--task_name $task_name \

--do_train \

--do_eval \

--max_seq_length 128 \

--per_device_train_batch_size 64 \

--per_device_eval_batch_size 32 \

--learning_rate 3e-4 \

--num_train_epochs 20 \

--output_dir $output_dir/model \

--overwrite_output_dir \

--logging_steps 1000 \

--logging_dir $output_dir/log \

--evaluation_strategy epoch \

--save_strategy epoch \

--warmup_ratio 0.06 \

--apply_expert_soup \

--adapter_size 16 \

--num_experts 4 \

--seed 0 \

--inference_level 3 \

--weight_decay 0.1 \

--sharing_up 1 \

--sharing_down 0 \

--use_consistency_loss 1

```

Most arguments are inherited from transformers and are easy to understand. We further explain some of the AdaMix's arguments:

* `inference_level`: There are two suggested modes

  * `1`: Random Routing

  * `3`: Averaging the weights of Adapters for routing (used in AdaMix)

* `num_experts`: Number of Adapters in AdaMix

* `use_consistency_loss`: Two modes. 

  * `0`: No consistency loss

  * `1`: Use consistency loss

* `sharing_up`: There are two modes. (sharing_down is same)

  * `0`: No weight sharing

  * `1`: Sharing Project-up layer weights in Adapter

### Evaluate the checkpoints

Create checkpoints directory and download checkpoints of corresponding tasks under the directory. Use MNLI as an example. Use your checkpoint path in **expert_soup_path** argument.

```console

export num_gpus=1

export PYTHONHASHSEED=0

task_name=mnli

model=roberta-large

export output_dir="./models/${model}/${task_name}"

python -m torch.distributed.launch --nproc_per_node=$num_gpus \

examples/text-classification/run_glue.py \

--model_name_or_path $model \

--task_name $task_name \

--do_eval \

--expert_soup_path ./checkpoints/pytorch_model_${task_name}_expert_soup.bin \

--max_seq_length 128 \

--per_device_train_batch_size 64 \

--per_device_eval_batch_size 32 \

--learning_rate 3e-4 \

--num_train_epochs 20 \

--output_dir $output_dir/model \

--overwrite_output_dir \

--logging_steps 1000 \

--logging_dir $output_dir/log \

--evaluation_strategy epoch \

--save_strategy epoch \

--warmup_ratio 0.06 \

--apply_expert_soup \

--adapter_size 16 \

--num_experts 4 \

--seed 0 \

--inference_level 3 \

--weight_decay 0.1 \

--sharing_up 1 \

--sharing_down 0 \

--use_consistency_loss 1

```

## Notes and Acknowledgments

The implementation is based on https://github.com/huggingface/transformers  


We also used some code from: https://github.com/microsoft/LoRA 

## Contact Information

For personal communication related to this package, please contact [Yaqing Wang](https://yaqingwang.github.io/) ([email protected]), [Sahaj Agarwal](https://www.linkedin.com/in/sahaj-agarwal-89aa49174/) ([email protected]), [Subhabrata (Subho) Mukherjee](https://www.microsoft.com/en-us/research/people/submukhe/) ([email protected]) or [Xiaodong Liu](https://sites.google.com/view/buptxiaodong/home) ([email protected]).

## How do I cite AdaMix?

```

@article{wang2022adamix,

  title={AdaMix: Mixture-of-Adaptations for Parameter-efficient Model Tuning},

  author={Wang, Yaqing and Agarwal, Sahaj and Mukherjee, Subhabrata and Liu, Xiaodong and Gao, Jing and Awadallah, Ahmed Hassan and Gao, Jianfeng},

  journal={arXiv preprint arXiv:2205.12410},

  year={2022}

}

```