HeadWiseKV allocates static per-head KV cache budgets using a layer-conditioned calibration algorithm, reducing memory by 8.59% at 112K context and extending verified context from 114K to 161K on Qwen3.6-27B.
Model Tells You What to Discard: Adaptive
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
years
2026 3representative citing papers
ScaleSearch optimizes block floating point scales via fine-grained search to cut quantization error by 27% for NVFP4, improving PTQ by up to 15 points on MATH500 for Qwen3-8B and attention PPL by 0.77 on Llama 3.1 70B.
DASH reduces prefill computation in LLMs and vision-language models by halting tokens with low attention-branch residual updates at a chosen layer, achieving 1.7–2× prefill speedups with minimal accuracy loss.
citing papers explorer
-
Search Your Block Floating Point Scales!
ScaleSearch optimizes block floating point scales via fine-grained search to cut quantization error by 27% for NVFP4, improving PTQ by up to 15 points on MATH500 for Qwen3-8B and attention PPL by 0.77 on Llama 3.1 70B.