Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Dialogue Without Limits: Constant-Sized KV Caches for Extended Responses in LLMs

About

Autoregressive Transformers rely on Key-Value (KV) caching to accelerate inference. However, the linear growth of the KV cache with context length leads to excessive memory consumption and bandwidth constraints. This bottleneck is particularly problematic in real-time applications -- such as chatbots and interactive assistants -- where low latency and high memory efficiency are critical. Existing methods drop distant tokens or compress states in a lossy manner, sacrificing accuracy by discarding vital context or introducing bias. We propose MorphKV, an inference-time technique that maintains a constant-sized KV cache while preserving accuracy. MorphKV balances long-range dependencies and local coherence during text generation. It eliminates early-token bias while retaining high-fidelity context by adaptively ranking tokens through correlation-aware selection. Unlike heuristic retention or lossy compression, MorphKV iteratively refines the KV cache via lightweight updates guided by attention patterns of recent tokens. This approach captures inter-token correlation with greater accuracy, crucial for tasks like content creation and code generation. Our studies on long-response tasks show 52.9$\%$ memory savings and 18.2$\%$ higher accuracy on average compared to state-of-the-art prior works, enabling efficient real-world deployment.

Ravi Ghadia, Avinash Kumar, Gaurav Jain, Prashant Nair, Poulami Das• 2025

Related benchmarks

TaskDatasetResultRank
Long-context UnderstandingLongBench 1.0 (test)
NarrativeQA30.35
108
Long-context retrieval accuracyRULER (test)
Accuracy (4K)84.62
24
Agentic CodingSWE-bench 50 tasks sampled
Resolved Count9
20
Showing 3 of 3 rows

Other info

Follow for update