header-langage
简体中文
繁體中文
English
Tiếng Việt
한국어
日本語
ภาษาไทย
Türkçe
Scan to Download the APP

Ramp Labs Introduces Multi-Agent Memory Sharing Solution, Token Consumption Reduced by Up to 65%

BlockBeats News, April 11th, AI infrastructure company Ramp Labs released research results on "Latent Briefing", achieving efficient memory sharing among multi-agent systems through direct compression of large-scale model KV cache, significantly reducing Token consumption without sacrificing accuracy.


In mainstream multi-agent architectures, the Orchestrator decomposes tasks and repeatedly calls Worker models. As the inference chain extends, Token usage exponentially inflates. The core idea of Latent Briefing is to leverage the attention mechanism to identify the truly critical parts in the context, directly discard redundant information at the representation layer, rather than relying on the slow-speed LLM summary or the unstable RAG retrieval.


In the LongBench v2 benchmark test, this method performed remarkably: Worker model Token consumption decreased by 65%, the median Token savings for medium-length documents (32k to 100k) reached 49%, the overall accuracy improved by approximately 3 percentage points compared to the baseline, and the additional time for each compression was only about 1.7 seconds, achieving a speedup of about 20 times compared to the original algorithm.


The experiment used Claude Sonnet 4 as the Orchestrator, and Qwen3-14B as the Worker model, covering various document scenarios such as academic papers, legal documents, novels, and government reports. The research also found that the optimal compression threshold varies depending on task difficulty and document length—difficult tasks are suitable for aggressive compression to filter out speculative reasoning noise, while long documents are more suitable for mild compression to retain scattered key information.

举报 Correction/Report
Correction/Report
Submit
Add Library
Visible to myself only
Public
Save
Choose Library
Add Library
Cancel
Finish