# Autoregressive architecture

Janus-Pro is a novel autoregressive framework that unifies multi-modal understanding and generation tasks and enhances flexibility by decoupling visual encoding.

Code Llama is a series of pre-trained and fine-tuned generative text models released by Meta with parameter scales ranging from 7 billion to 34 billion. This model is the 13-billion-parameter Python specialized version

Code Llama is a 34-billion-parameter Python-specific code generation model developed by Meta, optimized based on the Llama 2 architecture, focusing on Python code synthesis and understanding

GPT-Neo 2.7B is a 2.7 billion parameter Transformer language model replicated by EleutherAI based on the GPT-3 architecture, trained on the Pile dataset