Fengshenbang

library model
IDEA Lab
Open-source large model ecosystem from IDEA-CCNL covering NLU, NLG, and cross-modal tasks. Includes three model families: Erlangshen (NLU, largest public BERT at release, SOTA on FewCLUE/ZeroCLUE), Wenzhong (NLG, GPT-2 based generation), and Taiyi (cross-modal including first Chinese Stable Diffusion). All models available on HuggingFace.
GitHubHuggingFace

Outputs 3

Fengshenbang-LM

library

Open-source framework and model zoo for Chinese AI foundation models with Apache 2.0 license.

GitHubHuggingFace

GitHub Repository

Erlangshen

model

NLU model series including BERT, RoBERTa, and MegatronBERT variants up to 1.3B parameters. SOTA on FewCLUE and ZeroCLUE benchmarks.

HuggingFace

Wenzhong

model

Chinese NLG model series based on GPT-2 architecture with variants from 110M to 3.5B parameters.

HuggingFace
nlpopen-source