Q-Filters: Leveraging QK Geometry for Efficient KV Cache Compression
This addresses memory efficiency for large language models with long contexts, offering a practical compression solution that is incremental but competitive with existing methods.
The paper tackles the memory bottleneck of KV Cache in autoregressive language models by proposing Q-Filters, a training-free compression method that filters less crucial Key-Value pairs using Query-Key geometry, achieving 99% accuracy in retrieval with 32x compression and reducing perplexity drop by up to 65% in text generation compared to baselines.
Autoregressive language models rely on a Key-Value (KV) Cache, which avoids re-computing past hidden states during generation, making it faster. As model sizes and context lengths grow, the KV Cache becomes a significant memory bottleneck, which calls for compression methods that limit its size during generation. In this paper, we discover surprising properties of Query (Q) and Key (K) vectors that allow us to efficiently approximate attention scores without computing the attention maps. We propose Q-Filters, a training-free KV Cache compression method that filters out less crucial Key-Value pairs based on a single context-agnostic projection. Contrarily to many alternatives, Q-Filters is compatible with FlashAttention, as it does not require direct access to attention weights. Experimental results in long-context settings demonstrate that Q-Filters is competitive with attention-based compression methods such as SnapKV in retrieval tasks while consistently outperforming efficient compression schemes such as Streaming-LLM in generation setups. Notably, Q-Filters achieves a 99% accuracy in the needle-in-a-haystack task with a x32 compression level while reducing the generation perplexity drop by up to 65% in text generation compared to Streaming-LLM.