In this way, the input of LLM does not require visual tokens, which reduces the length of the input sequence and greatly improves efficiency. Following this paradigm, we propose VLoRA with the perceptual weights generator.

Tokenizer, serving as a translator to map the intricate visual data into a compact latent space, lies at the core of visual generative models.

Recently, various strategies for distributed training of large language models (LLMs) have been proposed.