How does the scaling behavior of inference throughput and reasoning accuracy differ between SMoES MoE-VLMs and

Fuente: Zenodo
Gespeichert in:
Bibliographische Detailangaben
1. Verfasser: SOVEREIGN Research Kernel
Format: Recurso digital
Sprache:Englisch
Veröffentlicht: Zenodo 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866902277238816768
author SOVEREIGN Research Kernel
author_facet SOVEREIGN Research Kernel
contents <p>In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve nea</p><p><strong>Research goal:</strong> How does the scaling behavior of inference throughput and reasoning accuracy differ between SMoES MoE-VLMs and dense counterparts when increasing total parameters from 7B to 34B on multimodal benchmarks (e.g., MMBench, SEED-Bench), and does the throughput gap widen or narrow at larger scales?</p><p><em>Autonomous synthesis report generated by SOVEREIGN Research Kernel. Tribunal consensus score: 7.7/10.</em></p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_20432231
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle How does the scaling behavior of inference throughput and reasoning accuracy differ between SMoES MoE-VLMs and
SOVEREIGN Research Kernel
scaling
behavior
inference
throughput
reasoning
accuracy
differ
SMoES
<p>In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve nea</p><p><strong>Research goal:</strong> How does the scaling behavior of inference throughput and reasoning accuracy differ between SMoES MoE-VLMs and dense counterparts when increasing total parameters from 7B to 34B on multimodal benchmarks (e.g., MMBench, SEED-Bench), and does the throughput gap widen or narrow at larger scales?</p><p><em>Autonomous synthesis report generated by SOVEREIGN Research Kernel. Tribunal consensus score: 7.7/10.</em></p>
title How does the scaling behavior of inference throughput and reasoning accuracy differ between SMoES MoE-VLMs and
topic scaling
behavior
inference
throughput
reasoning
accuracy
differ
SMoES
url https://doi.org/10.5281/zenodo.20432231