{"slug": "prompt-compression-via-activation-aggregation", "title": "Prompt Compression via Activation Aggregation", "summary": "Researchers have developed a method to compress instruction prompts into a single activation vector for large language models, achieving under 2% accuracy loss compared to full prompt processing. The technique uses a learned weighted sum of activations extracted at an intermediate layer and injected at an early layer, reducing per-query computation for fixed prompts. The findings also reveal cross-layer compatibility and robust compression in LLM activation spaces.", "body_md": "arXiv:2607.08399v1 Announce Type: new\nAbstract: Large language models process prompts by propagating activations through dozens of layers before generating a response. We ask whether the task-relevant information contained in an instruction prompt can be compressed into a single activation vector and re-injected into the model, replacing the original token sequence? We show this is achievable using a learned weighted sum of activations extracted at an intermediate layer and injected at an early layer of the target LLM. The compressed vector preserves task-relevant information, incurring an accuracy drop of under $2\\%$ relative to full prompt processing. Beyond its practical implications, including reducing per-query computation for fixed instruction prompts without reprocessing the original token sequence, our analysis reveals structure in the activation space of LLMs: (i) mid-layer representations transfer meaningfully to early layers, suggesting a degree of cross-layer compatibility in how information is encoded; (ii) a single activation vector encodes a quantifiable and recoverable amount of semantic information; (iii) a weighted sum of activations is a robust representation compressor.", "url": "https://wpnews.pro/news/prompt-compression-via-activation-aggregation", "canonical_source": "https://www.machinebrief.com/news/prompt-compression-via-activation-aggregation-rdjz", "published_at": "2026-07-10 04:00:00+00:00", "updated_at": "2026-07-10 04:20:47.765557+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-research"], "entities": [], "alternates": {"html": "https://wpnews.pro/news/prompt-compression-via-activation-aggregation", "markdown": "https://wpnews.pro/news/prompt-compression-via-activation-aggregation.md", "text": "https://wpnews.pro/news/prompt-compression-via-activation-aggregation.txt", "jsonld": "https://wpnews.pro/news/prompt-compression-via-activation-aggregation.jsonld"}}