Text Generation
文本生成(Text Generation)是自然语言处理领域的一项任务,旨在通过算法生成与人类书写的文本难以区分的内容。该任务利用马尔可夫过程或深度生成模型如LSTM实现,近期最先进方法包括BART、GPT及基于GAN的方案。系统评估通常采用人工评分或METEOR、ROUGE、BLEU等自动评价指标。文本生成在对话系统、自动摘要、机器翻译等领域具有重要应用价值。
ADGEN
AI2 Reasoning Challenge (25-Shot)
AI2 Reasoning Challenge TR
Alpaca-Eval (PT)
AlpacaEval
ARC-Challenge (PT)
Assin2 RTE
Assin2 STS
BBH (3-Shot)
BLUEX (No Images)
CALAME-PT
Censorship (0-shot)
Chinese Poems
RankGAN
CMU-SE
STWGAN-GP
CNN/Daily Mail
PALM
COCO Captions
LeakGAN
CommonGen
UniLM
Creativity (0-shot)
CrimeStats
CSL
Czech restaurant information
DailyDialog
DART
Drop (3-Shot)
EMNLP2017 WMT
LeakGAN
ENEM Challenge (No Images)
FaQuAD NLI
GPQA (0-shot)
GSM8k (5-shot)
GSM8k TR
HarmfulQA
GPT-4
HateBR Binary
HellaSwag (10-Shot)
HellaSwag (PT)
HellaSwag TR
Humanness (0-shot)
IFEval (0-Shot)
Internet
LAMBADA-PT
LCSTS
LDC2016E25
MATH Lvl 5 (4-Shot)
MMLU (5-Shot)
MMLU-PRO (5-shot)
MMLU TR
MT-Bench
MT-Bench-jp
MuSR (0-shot)
OAB Exams
One Billion Word
WGANGP + DGflow
Open-Mindedness (0-shot)
OpenWebText
PolContro
PT Hate Speech Binary
ReDial
UniCRS
ROCStories
Beam search + A*esque (sample)
SciQ
Stories/Jokes
Talking (0-shot)
TruthfulQA
TruthfulQA (0-shot)
TruthfulQA (PT)
TruthfulQA TR
tweetSentBR
Unruly
W/10
WikiText-103
Winogrande (5-shot)
Winogrande TR
World Knowledge (0-shot)
Yahoo Questions
Aggressive VAE