OmniScope: Modality-Decoupled Token Compression for Omnimodal Large Language Models
- 类型：arxiv
- 标识：2607.23193
- 链接：https://arxiv.org/abs/2607.23193
- 主分类：multimodal
- 形态：method
- 被引：0
- 被引来源：Semantic Scholar + OpenAlex
- S2被引：0
- OpenAlex被引：0
- 影响力被引：0
- TLDR：This work proposes OmniScope, a training-free token compression framework that uses the query as a shared semantic anchor while estimating relevance separately for audio and video, and suggests a simple design principle for OmniLLM inference: share the query across modalities, but not the salience estimates.
- OpenAlex ID：W7171563028
- OpenAlex DOI：10.48550/arxiv.2607.23193
- DOI：10.48550/arxiv.2607.23193
- DOI来源：OpenAlex
- 开放获取：green
- 开放获取链接：https://doi.org/10.48550/arxiv.2607.23193
- OpenAlex更新：2026-08-26
- 待LLM分类：否
- 标题中文：OmniScope：面向全模态大语言模型的模态解耦 token 压缩
- TLDR中文：提出 OmniScope，一个无需训练的 token 压缩框架，以 query 作为跨模态共享的语义锚点，并对音频与视频分别估计相关性；由此给出 OmniLLM 推理的简单设计原则：跨模态共享 query，但不共享显著性估计。
- 来源文件：
- /inbox/tom/_candidates/2026-08-01-agent-rag-longcontext-candidates.json
- /inbox/tom/_candidates/2026-08-02-agent-rag-longcontext-candidates.json
- /inbox/tom/_candidates/2026-08-03-agent-rag-longcontext-candidates.json
- /inbox/tom/_candidates/2026-08-03-agent-memory-tool-use-candidates.json
- [S2 enrich]
- [OpenAlex backfill]