Seed based weight compression improves security and efficiency in large language models
Security-Enhanced Seed-Based Weight Quantization for Large Language Models
Cryptography and SecurityArtificial IntelligenceMachine Learning
Summary
Large language models need a lot of storage and energy, so finding ways to shrink them is important. The authors propose Seed-Q, a method that better compresses model data by paying more attention to important parts of the model. This method also makes it easier to spot if someone tampers with the model’s data. They show their approach works well across different models and can be used with special hardware.
What this means in practice
- •For machine learning engineers: Use Seed-Q to compress large language models efficiently while preserving performance and improving resistance to data corruption.
- •For hardware accelerator designers: Implement Seed-Q’s compression scheme in ASICs to reduce model storage needs and add security with minimal hardware overhead.
Authors
Qiuyu Ren, Sudipta Paria, Aritra Dasgupta, Swarup Bhunia
Abstract
Large language models (LLMs) incur substantial storage, memory-bandwidth and energy costs, motivating compact weight representations. Existing seed-based compression methods reconstruct weights from compact pseudo-random representations but do not explicitly account for the non-uniform sensitivity of model weights. We introduce Seed-Q, a security-enhanced sensitivity-aware seed-based weight compression framework that uses lightweight Linear Feedback Shift Register (LFSR)-based weight generation with non-uniform bit allocation. Our approach assigns larger representation budgets to sensitive weights while aggressively compressing less sensitive regions. Importantly, this non-uniform allocation requires no side-information: the decoder deterministically reconstructs the bit-allocation schedule, with no rung depending on the decoded weights, eliminating the need to store per-block metadata or use calibration data while preserving the baseline coding rate. Experiments across diverse LLMs show that Seed-Q matches 4-bit perplexity of SeedLM with fewer bits, while at the same 4 bits/weight it reduces both perplexity degradation and zero-shot accuracy loss relative to SeedLM. We also show that Seed-Q simultaneously achieves high security against bit-flip attacks on model parameters, as bit corruption affects multiple reconstructed weights, greatly amplifying its impact and making it easier to detect. We further implement Seed-Q in an ASIC-based accelerator and demonstrate modest hardware overhead compared to prior seed-based approaches.