✨OmniZip: Audio-Guided Dynamic Token Compression for Fast Omnimodal Large Language Models

📝 Summary:
OmniZip is a training-free framework that addresses the computational bottleneck in omnimodal LLMs by dynamically compressing audio-visual tokens. It uses audio retention scores to guide video token pruning, achieving 3.42X inference speedup and 1.4X memory reduction without performance loss.

🔹 Publication Date: Published on Nov 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.14582
• PDF: https://arxiv.org/pdf/2511.14582
• Github: https://github.com/KD-TAO/OmniZip

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#OmnimodalLLM #TokenCompression #LLMs #AI #ModelEfficiency

✨Detecting Overflow in Compressed Token Representations for Retrieval-Augmented Generation

📝 Summary:
Soft compression for LLMs can lead to token overflow, losing vital information. This paper proposes query-aware probing classifiers that detect overflow with 0.72 AUC-ROC, improving upon query-agnostic methods. This enables low-cost pre-LLM gating to mitigate compression-induced errors.

🔹 Publication Date: Published on Feb 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.12235
• PDF: https://arxiv.org/pdf/2602.12235

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#LLMs #RAG #NLP #AIResearch #TokenCompression