Skip to content

Latest commit

 

History

History
60 lines (39 loc) · 1.43 KB

README.md

File metadata and controls

60 lines (39 loc) · 1.43 KB

Merging Language Models (LMs)

Get Started

Dependencies

Please follow DARE to install the dependencies.

Additionally, install scipy, sklearn, torchmetrics, evaluate.

Checkpoints

RoBERTa: You can download the fine-tuned checkpoints from huggingface here.

GPT-2: You can download the fine-tuned checkpoints from huggingface here.

Place the checkpoints as follows:

│cktps/
├──roberta/
│  ├── cola/
│  │  ├── roberta-base_lr1e-05
│  │  │  ├── config.json
│  │  │  ├──......
│  ├── sst2/
│  │  ├── roberta-base_lr1e-05
│  │  │  ├── config.json
│  │  │  ├──......
│  ├── ......
├──gpt2/
│  ├── gpt2/
│  │  ├── config.json
│  │  ├──......
│  ├── gpt2_cola/
│  │  ├── config.json
│  │  ├──......
│  ├── ......

You can modify the cache_dir in the utils/load_config.py file to specify your own path to save the datasets.

Eval

Merge RoBERTa models

python merge_roberta_glue.py

Merge GPT-2 models

python merge_gpt_glue.py

Results

Results for our EMR-Merging will be saved in ./save_merge_logs.