LLM Reference

Yi-1.5 34B

About

Yi-1.5-34B, developed by 01.AI, is an advanced large language model designed to function as an adept chatbot. As an enhancement of its predecessors, it excels in natural language understanding, coding, mathematics, and reasoning. It is built on a transformer architecture incorporating innovative features like Grouped-Query Attention and SwiGLU activation to efficiently process long contexts of up to 200,000 tokens. Trained on a vast corpus consisting of both Chinese and English texts, the model has been fine-tuned with a curated instruction dataset to boost performance and reliability. While it often matches or surpasses larger models in various benchmarks, it is also available for deployment via platforms like Hugging Face and utilizes quantization techniques for resource efficiency. However, users should be aware of potential hallucinations and biases due to training data limitations.

Capabilities

MultimodalFunction CallingTool UseJSON Mode

Specifications

Released2024-05-12
Parameters34B
ArchitectureDecoder Only
Specializationgeneral