SuperCLUE uses the Elo rating system to calculate the relative performance of models. SuperCLUE Nirvana in Fire will release new rankings regularly.
Last update: 2023-05-29 18:22:35
1、 anthropic-claude
由 Anthropic 开发,基于 Anthropic 对训练有用、诚实、无害的人工智能系统的研究。
2、MiniMax-abab5
由 MINIMAX 公司开发,帮助人类高效写作、激发创意、获取知识、做出决策。
3、gpt-3.5-turbo
由 OpenAI 开发,当前大家最频繁使用的版本。
4、ChatGLM-130B
Developed by Tsinghua University and Zhipu AI, it is an open bilingual dialogue model.
5、ChatGLM-6B
Developed by Tsinghua University and Zhipu AI, it is an open source bilingual dialogue model.
6、 phoenix-inst-chat-7b
Developed by The Chinese University of Hong Kong (Shenzhen), a multilingual chat assistant based on fine-tuning of Bloomz.
7、 moss-moon-003-sft
Developed by the Natural Language Processing Laboratory of Fudan University, it is an open-source dialogue language model that supports Chinese-English bilingual and various plug-ins.
8. Longjing-7B
New branch Large Language Models under ChatYuan.
9. idea-jiangziya
Developed by IDEA Research Institute CCNL, it rebuilds the Chinese vocabulary from LLaMA-13B, and performs pre-training on the scale of 100 billion tokens, so that the model has native Chinese capabilities.
10、vicuna-13b
Developed by LMSYS, based on LLaMA and fine-tuned using user-shared dialogue.
11、Belle-13B
Developed by LianjiaTech, optimized for Chinese based on BLOOM and LLAMA, model tuning only uses data generated by ChatGPT to provide better support for Chinese instructions.
12、LMFlow-Robin-7B
Developed by the HKUST Statistics and Machine Learning Laboratory team, the model is a fine-tuned version of pinkmanlove/llama-7b-hf on a custom dataset.
13、Linly-ChatFlow-7B
Developed by the National Engineering Laboratory of Big Data System Computing Technology, it is obtained by fine-tuning the Chinese basic model on the 4 million instruction data set
14、RWKV-4-Raven-7B
Developed by the RWKV Foundation, it is a language model that combines RNN and Transformer. It is suitable for long texts, runs faster, has better fitting performance, occupies less video memory, and takes less training time.
15、Chinese-Alpaca-Plus-13B
A version of the Chinese Alpaca model, on the basis of the original LLaMA model, expands the Chinese vocabulary and uses Chinese data for secondary pre-training, further improving the basic semantic understanding of Chinese.
16、Bloomz-7b1-mt
BLOOMZ & mT0 series from bigscience, a family of models capable of executing human instructions in dozens of languages.
Check out the latest rankings on the official website of SuperCLUE Nirvana Rankings ;