LLaVA-Gemma: Accelerating Multimodal Foundation Models with a Compact Language Model
This work addresses the development of compact multimodal models for AI applications, but it is incremental as it replicates existing methods without achieving new benchmarks.
The authors trained multimodal foundation models using the LLaVA framework with Gemma language models, focusing on a 2B parameter version, and found that ablating design features like pretraining and model sizes led to moderate performance but no improvements over current state-of-the-art models of similar scale.
We train a suite of multimodal foundation models (MMFM) using the popular LLaVA framework with the recently released Gemma family of large language models (LLMs). Of particular interest is the 2B parameter Gemma model, which provides opportunities to construct capable small-scale MMFMs. In line with findings from other papers in this space, we test the effect of ablating three design features: pretraining the connector, utilizing a more powerful image backbone, and increasing the size of the language backbone. The resulting models, which we call LLaVA-Gemma, exhibit moderate performance on an array of evaluations, but fail to improve past the current comparably sized SOTA models. Closer analysis of performance shows mixed effects; skipping pretraining tends to reduce performance, larger vision models sometimes improve performance, and increasing language model size has inconsistent effects. We publicly release training recipes, code and weights for our models for the LLaVA-Gemma models.