Zero-Shot Vision Encoder Grafting via LLM Surrogates

TL;DR

This paper introduces a zero-shot grafting method that trains small surrogate vision models to efficiently transfer to large language models, significantly reducing training costs while maintaining high performance.

Contribution

The authors propose a novel zero-shot grafting technique using surrogate models to transfer vision encoders to large LLMs, reducing training costs by 45%.

Findings

Surrogate models share the same embedding space as large LLMs.

Grafted models outperform surrogate-only models and match full training performance.

Training costs are reduced by approximately 45% with the proposed method.

Abstract

Vision language models (VLMs) typically pair a modestly sized vision encoder with a large language model (LLM), e.g., Llama-70B, making the decoder the primary computational burden during training. To reduce costs, a potential promising strategy is to first train the vision encoder using a small language model before transferring it to the large one. We construct small "surrogate models" that share the same embedding space and representation language as the large target LLM by directly inheriting its shallow layers. Vision encoders trained on the surrogate can then be directly transferred to the larger model, a process we call zero-shot grafting -- when plugged directly into the full-size target LLM, the grafted pair surpasses the encoder-surrogate pair and, on some benchmarks, even performs on par with full decoder training with the target LLM. Furthermore, our surrogate training approach reduces overall VLM training costs by ~45% when using Llama-70B as the decoder. The code is at https://github.com/facebookresearch/zero.