Skip to content

Latest commit

 

History

History
25 lines (14 loc) · 871 Bytes

README.md

File metadata and controls

25 lines (14 loc) · 871 Bytes
Raw

Sequence_Span_Rewriting

Code for EMNLP 2021 paper Improving Sequence-to-Sequence Pre-training via Sequence Span Rewriting

Usage

data_generation.py contains key functions of generating training data for the sequence span rewriting objective.

data_gen.py contains an example of data generation.

run_summarization.py is from Huggingface Transformers. We use this file to continually per-train with SSR and fine-tune it on downstream tasks.

run_generation.py is used for inference (i.e., generation).

Pre-trained models

You can load our pre-trained SSR-base from Huggingface's model hub:

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
  
tokenizer = AutoTokenizer.from_pretrained("microsoft/ssr-base")

model = AutoModelForSeq2SeqLM.from_pretrained("microsoft/ssr-base")