History

DingDing 0406866c25 Merge branch 'thunlp:main' into main		2022-03-13 22:12:26 +08:00
..
examples_bmtrain	add bmtrain	2022-03-13 01:21:55 +08:00
examples_image-classification	Update README.md	2022-02-14 23:28:40 +08:00
examples_prompt	add bmtrain	2022-03-13 01:21:55 +08:00
examples_seq2seq	remove tfevent	2022-03-13 01:23:54 +08:00
examples_text-classification	Merge branch 'thunlp:main' into main	2022-03-13 22:12:26 +08:00
tutorial	lora	2022-02-20 17:24:59 +08:00
README.md	first commit	2022-02-14 21:19:03 +08:00
setup_prompt.py	add example	2022-02-26 14:24:42 +08:00
setup_seq2seq.py	first commit	2022-02-14 21:19:03 +08:00

README.md

Use Examples

This repo mainly contains several running scripts to use OpenDelta to conduct parameter-efficient training of various tasks.

Note that we suggest adding OpenDelta to existing scripts, instead of modify a scripts into the following examples. OpenDelta itself doens't restrict the training pipeline nor provide pipeline.

tutorial

Several toy tutorials:

The scripts for docs/basic_usage
Using interactive module selection
Work with OpenPrompt

examples_text-classification

Modify a huggingface text-classification examples into a delta tuning one. Currently, GLUE datasets are supported in the scripts. Roberta-base is used for performance checking. Read README.md inside the repo for detailed usage.

examples_seq2seq

Modify a huggingface sequence to sequence examples into a delta tuning one. Currently, SuperGLUE and GLUE datasets are supported in the scripts. T5-base is used for performance checking. Read README.md inside the repo for detailed usage.

examples_image-classification

A toy example of using OpenDelta for a Computer Vision Pretrained Model (ViT). Since ViT is an experimental feature in huggingface transformers, this example is subject to Change at any moment.