ICTNLP/bayling-7b-diff

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kLicense:gpl-3.0Architecture:Transformer0.0K Open Weights Cold

ICTNLP/bayling-7b-diff is a 7 billion parameter instruction-following large language model developed by the NLP Group of Institute of Computing Technology, Chinese Academy of Sciences (ICT/CAS). Optimized for cross-lingual alignment and interactive translation, BayLing excels in English and Chinese generation, instruction following, and multi-turn interactions. This model is a weight-diff version of BayLing-7B, designed for efficient deployment on consumer-grade GPUs with 16GB memory. Its primary strength lies in bridging language barriers for tasks like translation, writing, and content creation.

Loading preview...

BayLing: Cross-lingual Instruction-Following LLM

BayLing (百聆) is a 7 billion parameter instruction-following large language model developed by the NLP Group of ICT/CAS, specifically designed to enhance cross-lingual alignment and instruction following through interactive translation. This model demonstrates strong capabilities in both English and Chinese generation, multi-turn interaction, and adhering to instructions.

Key Capabilities & Features

  • Multilingual Proficiency: Excels in English and Chinese generation and instruction following.
  • Interactive Translation: Bridging cross-lingual alignment through interactive translation.
  • Efficient Deployment: The weight-diff version of BayLing-7B can be deployed on consumer-grade GPUs with 16GB of memory.
  • Instruction Following: Strong performance in understanding and executing multi-turn instructions.
  • Evaluation Set: Accompanied by the BayLing-80 Test Set, a human-annotated evaluation set for multilingual and multi-turn interaction capabilities.

Limitations

BayLing, like other LLMs, has limitations. It may generate inaccurate factual information and lacks proficiency in reasoning, mathematics, and coding tasks. There is also a risk of generating harmful or biased content. Users should be aware that the model cannot guarantee absolute accuracy of generated content.