# SPHINX-X: Scaling Data and Parameters for a Family of Multi-modal Large Language Models > Research note on SPHINX-X, an enhanced and scaled version of the SPHINX multimodal LLM with simplified training and improved efficiency. - Authors: Hunter Heidenreich - Published: 2024-02-08 - Updated: 2026-03-04 - Paper: SPHINX-X Paper - Code: SPHINX-X GitHub - Source: Bevaya Labs, --- ## What kind of paper is this? - Small idea paper - This is an introduction of some enhancements made to the [SPHINX model](2023_11_sphinx.md) and evaluation of the model's performance across various vision-language tasks to demonstrate the improvements. ## What is the motivation? - The motivation is to **scale up** the SPHINX model, make it easier to train, and to simplify its complexity while speeding up training and inference - The potential impact is to make the SPHINX model more accessible and efficient for researchers and practitioners working on vision-language tasks. ## What is the novelty? - Related work includes the original SPHINX model and other multi-modal models like LLaVA, MiniGPT-4, and BLIP-2. - Very minor contributions from this paper, detailed below. ## What experiments were performed? - Better performance on various vision-language tasks compared to the original SPHINX model. ## Dataset ## Methodology Highlights - Two of the four original vision encoders are dropped in favor of the two complementary and better-performing encoders. - They use a trick to skip visual tokens constructed from entirely padded values; this is only really relevant for wide or tall aspect ratio images. This involves a learned skip token which can be used to replace the padding tokens, tightening up the sequence length while still preserving relative positioning. - They ditch a two-stage training pipeline in favor of a single stage. - Other changes are really just including more and different data, other models as LLM backbones, etc. ## Training Details