Back to feed
arXiv cs.CL
arXiv cs.CL
6/23/2026
Less is More: Lightweight Prompt Compression for Question Answering Applications on Edge Devices

Less is More: Lightweight Prompt Compression for Question Answering Applications on Edge Devices

Short summary

CORE is a lightweight prompt compression method for RAG-based QA systems that eliminates auxiliary language models, using named entity recognition and semantic matching. It achieves 30% accuracy gains and 50%+ memory reduction on edge devices like smartphones and Jetson boards. Energy consumption drops 96% compared to existing methods, making it practical for mobile deployment.

  • •Two-stage compression using NER and semantic matching—no auxiliary models needed
  • •30% accuracy improvement, 50%+ memory reduction, 2x speedup on edge devices
  • •96% energy reduction on smartphones vs. LLMLingua2

Generated with AI, which can make mistakes.

#ai-tools#research-breakthrough#ai-agents
Read full article at arXiv cs.CL

Is this a good recommendation for you?

Explore more

Closing the Social-Semantic Gap: SPSD for Edge-Based Prompt Compression in Cloud LLM Inference
arXiv cs.LG
Context Compression Is Not One Thing: Readable Symbolic Re-expression vs. Coherent Summary at Matched Budget
arXiv cs.CL
SproutRAG: Attention-Guided Tree Search with Progressive Embeddings for Long-Document RAG
arXiv cs.CL
MODE: Modality-Decomposed Expert-Level Mixed-Precision Quantization for MoE Multimodal LLMs
arXiv cs.LG