A JoLT for the KV cache: Near-Lossless KV Cache Compression via Joint Rank-bit Allocation
Abstract
The key-value (KV) cache is the dominant memory bottleneck in long-context language model inference. Existing compression methods apply low-rank factorization or quantization independently, without jointly allocating rank and precision under a shared storage budget. We introduce JoLT, a training-free compressor that treats grouped prefill caches as fourth-order tensors and applies partial Tucker decomposition along the token and feature modes, the two axes that carry low-rank structure, while leaving the head and layer modes intact. A rotated low-bit quantizer captures the truncation residual, and a single Lagrangian dual allocates per-group Tucker ranks and residual bit-widths under a global byte constraint. FlashJoLT replaces the exact token-mode SVD with a randomized approximation that matches JoLT within the free zone at a fraction of the compression cost, and a fused Triton decode kernel evaluates attention directly over the stored factors without materializing dense KV tensors. Across five models from four architecture families, covering multi-head attention, grouped-query attention, and mixture-of-experts architecture, JoLT achieves 2 - 3x compression with less than 0.2% perplexity degradation, without retraining. On RULER at 64K context with LLaMA-3.1-8B, retrieval accuracy remains near-lossless through 3x and declines by only 0.90 and 2.40pp at 4x and 5x, respectively. JoLT demonstrates that tensor-aware low-rank decomposition and quantized residuals, unified under a single storage budget, achieve near-lossless KV-cache compression across diverse model architectures without retraining.
Related
- DiffLyap-MEC: Diffusion-Guided Lyapunov Optimization for Joint Offloading, Semantic Compression, and Resource Allocation in ISAC-Enabled 6G Edge Computing
- A Token/KV-Cache Communication Media Selection and Resource Allocation Strategy for Multi-Agent Collaboration
- Finite-Bit CSI Compression Under Feedback Constraints: Exact PCA Error Decomposition and Robust Dimension–Bit Allocation for MISO-OFDM
- Joint Quantized Precoding and Bit Allocation for Fronthaul-Constrained Cell-Free Massive MIMO
- Efficient Joint Trajectory Optimization and Resource Allocation for Secure UAV Networks: A Deep Reinforcement Learning Approach
BibTeX
@misc{krishnan2026jolt,
title = {{A JoLT for the KV cache: Near-Lossless KV Cache Compression via Joint Rank-bit Allocation}},
author = {Rahul Krishnan and Volker Schulz},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2607.12550},
eprint = {2607.12550},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2607.12550},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - A JoLT for the KV cache: Near-Lossless KV Cache Compression via Joint Rank-bit Allocation
AU - Rahul Krishnan
AU - Volker Schulz
PY - 2026
DA - 2026/09/24
PB - arXiv
DO - 10.48550/arxiv.2607.12550
UR - https://arxiv.org/abs/2607.12550
AB - The key-value (KV) cache is the dominant memory bottleneck in long-context language model inference. Existing compression methods apply low-rank factorization or quantization independently, without jointly allocating rank and precision under a shared storage budget. We introduce JoLT, a training-free compressor that treats grouped prefill caches as fourth-order tensors and applies partial Tucker decomposition along the token and feature modes, the two axes that carry low-rank structure, while leaving the head and layer modes intact. A rotated low-bit quantizer captures the truncation residual, and a single Lagrangian dual allocates per-group Tucker ranks and residual bit-widths under a global byte constraint. FlashJoLT replaces the exact token-mode SVD with a randomized approximation that matches JoLT within the free zone at a fraction of the compression cost, and a fused Triton decode kernel evaluates attention directly over the stored factors without materializing dense KV tensors. Across five models from four architecture families, covering multi-head attention, grouped-query attention, and mixture-of-experts architecture, JoLT achieves 2 - 3x compression with less than 0.2% perplexity degradation, without retraining. On RULER at 64K context with LLaMA-3.1-8B, retrieval accuracy remains near-lossless through 3x and declines by only 0.90 and 2.40pp at 4x and 5x, respectively. JoLT demonstrates that tensor-aware low-rank decomposition and quantized residuals, unified under a single storage budget, achieve near-lossless KV-cache compression across diverse model architectures without retraining.
ER -