EISL

Don’t Take It Literally: An Edit-Invariant Sequence Loss for Text Generation

Guangyi Liu, Zichao Yang, Tianhua Tao, Xiaodan Liang, Zhen Li, Bowen Zhou, Shuguang Cui, Zhiting Hu

Non-Autoregressive Machine Translation

NAT codes (fairseq) and models are released in the NAT folder. We also provide the models trained by CE loss. The results are reported in the paper.

Name		Name	Last commit message	Last commit date
Latest commit History 5 Commits
.ipynb_checkpoints		.ipynb_checkpoints
__pycache__		__pycache__
bpe		bpe
build		build
complexity_out		complexity_out
docs		docs
examples		examples
fairseq.egg-info		fairseq.egg-info
fairseq		fairseq
fairseq_cli		fairseq_cli
old_shell		old_shell
scripts		scripts
tests		tests
.gitignore		.gitignore
.gitignore.swo		.gitignore.swo
.gitmodules		.gitmodules
0.sh		0.sh
1.sh		1.sh
2.sh		2.sh
3.sh		3.sh
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
beam_bart_wmt.sh		beam_bart_wmt.sh
checkpoints		checkpoints
data-bin		data-bin
draw_loss.sh		draw_loss.sh
eval_P.sh		eval_P.sh
generate_noise.sh		generate_noise.sh
generate_noisy_MT.py		generate_noisy_MT.py
generate_noisy_wmt.py		generate_noisy_wmt.py
hubconf.py		hubconf.py
iclr_reb_trunction.sh		iclr_reb_trunction.sh
iclr_reb_trunction_pre.sh		iclr_reb_trunction_pre.sh
iclr_reb_trunction_pre_all.sh		iclr_reb_trunction_pre_all.sh
iclr_reb_trunction_pre_all_blank.sh		iclr_reb_trunction_pre_all_blank.sh
iclr_reb_trunction_pre_all_syn.sh		iclr_reb_trunction_pre_all_syn.sh
log		log
pre_data.sh		pre_data.sh
pre_data_bart.sh		pre_data_bart.sh
pre_data_bart_wmt.sh		pre_data_bart_wmt.sh
pyproject.toml		pyproject.toml
raw_wmt.sh		raw_wmt.sh
raw_wmt_1.sh		raw_wmt_1.sh
raw_wmt_2.sh		raw_wmt_2.sh
raw_wmt_3.sh		raw_wmt_3.sh
setup.py		setup.py
train.py		train.py