mlsanityData to 7 Oct 2026

IsoQuant: Hardware-Aligned SO(4) Isoclinic Rotations for LLM KV Cache Compression

arXiv 2603.28430 ↗First posted 30 Mar 2026cs.LG cs.CLcited by 2

On the map: Efficient LLM inference and serving → KV cache quantization

Abstract (arXiv)

Orthogonal feature decorrelation is effective for low-bit online vector quantization, but dense random orthogonal transforms incur prohibitive O(d2) storage and compute. RotorQuant reduces this cost with blockwise 3D Clifford rotors, yet the resulting 3D partition is poorly aligned with modern hardware and offers limited local mixing. We propose IsoQuant, a blockwise rotation framework based on quaternion algebra and the isoclinic decomposition of SO(4). It represents each 4D block as a quaternion and applies a closed-form transform T(v)=qL v overlineqR. This yields two main variants: IsoQuant-Full, which realizes the full SO(4) rotation, and IsoQuant-Fast, which keeps only one isoclinic factor for lower cost; the framework also admits a lightweight 2D special case. At d=128, IsoQuant-Full reduces forward rotation cost from about 2,408 FMAs in RotorQuant to 1,024, while IsoQuant-Fast further reduces it to 512. Across 18 fused CUDA settings with d ∈ 128,256,512, bit widths 2,3,4, and FP16/FP32 execution, IsoQuant achieves mean kernel-level speedups of about 4.5×--4.7× over RotorQuant while maintaining comparable reconstruction MSE, with peak speedups above 6×. Current validation is limited to the stage-1 quantize--dequantize path on synthetic normalized vectors; end-to-end KV-cache evaluation remains future work.

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in KV cache quantization, one of 12 directions in Efficient LLM inference and serving. Builds on 5 of its 15 references that are in the map. First posted on arXiv 30 Mar 2026.

Builds on

5 of 15 references.

Closest ancestors · 3

Same direction, newest first

  1. 2024QJL: 1-Bit Quantized JL Transform for KV Cache Quantization with Zero Overhead5 Jun
  2. 2024KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache5 Feb
  3. 2024KVQuant: Towards 10 Million Context Length LLM Inference with KV Cache Quantization31 Jan

Other references · 2

  1. 2025TurboQuant: Online Vector Quantization with Near-optimal Distortion Ratecited by 116
  2. 2025RiemannFormer: A Framework for Attention in Curved Spacescited by 8

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026OScaR: The Occam's Razor for Extreme KV Cache Quantization in LLMs and Beyondcited by 5
  2. 2026Hurwitz Quaternion Multiplicative Quantization for KV Cache Compressioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026OCTOPUS: Optimized KV Cache for Transformers via Octahedral Parametrization Under optimal Squared error quantization
  2. 2026TurboAngle: Near-Lossless KV Cache Compression via Uniform Angle Quantization
  3. 2026FibQuant: Universal Vector Quantization for Random-Access KV-Cache Compression
  4. 2026ResidualQuant: KV Cache Quantization for Looped Transformers with 2-Bit Residuals
  5. 2025PolarQuant: Leveraging Polar Transformation for Efficient Key Cache Quantization and Decoding Acceleration
  6. 2026OSCAR: Offline Spectral Covariance-Aware Rotation for 2-bit KV Cache Quantization
  7. 2024QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMs
  8. 2026Sequential KV Cache Compression via Probabilistic Language Tries: Beyond the Per-Vector Shannon Limit