This paper proposes a post-training quantization method called VLCQ that achieves competitive or improved performance while reducing bit-widths without requiring access to training data or model fine-…

AQ4SViT is a novel automated quantization framework for Spiking Vision Transformers that provides quick quantization settings with good trade-offs between accuracy and memory.

Derives the optimal quantizer of a real-valued random variable with a specified output distribution and minimum mean squared error.

The paper introduces Fast-TurboQuant, a multiplier-free projection architecture for large language models that uses a structured fast Johnson-Lindenstrauss transform instead of dense matrices, resulti…

This paper proposes KroQuant, a post-training quantization method for diffusion transformers using learned Kronecker-structured invertible transforms, which reduces parameters, improves speed, and mai…

This paper proposes a method for quantizing large language models with cross-layer error compensation and finite-sample feature-statistics matching.

This paper analyzes the effect of quantization on the Information Bottleneck framework and proposes task-oriented quantization designs for Gaussian and non-Gaussian settings.

This paper introduces a provably sound runtime meter for KV-cache quantization to ensure compression does not damage the served request in real-time.

This paper systematically analyzes combining dimensionality reduction and quantization to compress text embeddings, showing that this combined approach achieves substantial compression (e.g., 0.1% siz…

HARP introduces a novel, adaptive, learnable orthogonal processor that significantly improves the robustness and accuracy of extreme low-bit LLM quantization compared to fixed methods.

Establishes conditions for embedding documents as vectors for top-k retrieval and analyzes the required dimension and precision.

The paper introduces VibeVoice-ASR-BitNet, a compressed real-time speech recognition model optimized for edge CPUs using heterogeneous quantization and custom SIMD kernels.

This paper investigates the necessity of interaction for order-optimal 1-bit mean estimation in nonparametric finite-moment classes.

The paper proposes BAMU, a framework for dynamic RVQ allocation in pretrained neural speech codecs, achieving consistent improvements on EnCodec and DAC.

This paper proposes SpikeQ, a spike-guided post-training quantization framework for spiking neural networks.

This paper examines the use of 2D barcodes on temporary travel documents to enable machine readability and automate biometric face verification while reducing storage capacity. It compares the perform…