How does the inference throughput (tokens per second) of Uni-MoE-2.0-Omni's dynamic-capacity MoE compare to de

Abstract

Abstract: We present Uni-MoE 2.0 from the Lychee family. As a fully open-source omnimodal large model (OLM), it substantially advances Lychee's Uni-MoE series in language-centric multimodal understanding, reasoning, and generating. Based on the dense LLM, we build Uni-MoE-2.0-Omni from scratch through three core contributions: dynamic-capacity Mixture-of-Experts (MoE) design, a progressive training strategy enhanced with an iterative reinforcement strategy, and a carefully curated multimodal data matching technique. It is capable of omnimodal understanding, as well as generating images, text, and speech

Research Question

How does the inference throughput (tokens per second) of Uni-MoE-2.0-Omni's dynamic-capacity MoE compare to dense baselines and fixed-top-k hard-routed MoE on multimodal QA benchmarks (e.g., MMBench, MMMU) when varying the number of active experts from 2 to 8 under batch sizes of 1, 16, and 64?

Verification Level

Descriptive public verification status only; aggregate claim counts are public, but individual claim records are not exposed here.

Quality Tier

Descriptive public triage only; this tier does not alter current publication or DOI behavior.

Quality Dimensions

Automated triage signals derived from public fields; not human peer review or independent validation.

Correction Record

Public corrections are additive records. Current status does not claim the synthesis is error-free.

Provenance