LLM Reference

XuanYuan 70B

About

The XuanYuan 70B is a cutting-edge large language model by Du Xiaoman Financial, designed specifically for the financial industry. Built on the Llama2-70B architecture, it supports a context length of up to 16k tokens in some versions, providing robust handling of long and complex financial queries. This model was trained extensively on Chinese financial data, resulting in proficient text generation and financial analysis capabilities. It features efficient 8-bit and 4-bit quantization and achieves a remarkable training efficiency of 340 tokens/s/gpu on a 100-node cluster. Additionally, it offers a chat model version for improved conversational competence. However, it requires high-quality prompts and may reflect biases inherent in its training data. A more advanced version, XuanYuan 2.0, utilizes the BLOOM-176B architecture with hundreds of billions of parameters.

Capabilities

MultimodalFunction CallingTool UseJSON Mode

Specifications

FamilyXuanYuan
Released2024-02-03
Parameters70B
ArchitectureDecoder Only
Specializationgeneral