ThinQuant introduces efficient rotation learning for low‑bit weight and activation quantization of large language models by reducing calibration data through a geometric selection of activations and solving a lower‑dimensional optimization problem via an ADMM algorithm. The method achieves comparable or better quantization performance with dramatically fewer calibration points, completing rotation calibration for Llama‑3‑70B in under 12 minutes and for Llama‑3.1‑405B in just over 2 hours on a single GPU. ThinQuant outperforms existing gradient‑free approaches such as DartQuant and gradient‑based SpinQuant in both speed and perplexity metrics on WikiText‑2.
arXiv:2605.11222v2 Announce Type: replace Abstract: Quantization is an effective strategy to reduce the storage and computation footprint of large language models (LLMs). Post-training quantization (...
HARP (Hadamard‑Preconditioned Adaptive Rotation Processor) is a learnable, structured two‑sided orthogonal processor that replaces fixed randomized Hadamard transforms in post‑training quantization of large language models. By representing rotations as sparse butterfly‑like block‑orthogonal stages and supporting mixed‑radix schedules, HARP adapts the quantization basis to each layer and calibration distribution while maintaining full‑precision equivalence. Across 2–4‑bit settings on Llama models from 1B to 70B, HARP consistently improves perplexity, delivers the strongest zero‑shot gains at 2 bits, and preserves deployment efficiency—achieving 128 tokens per second on Llama 2 7B at 2 bits, roughly 90% of RHT throughput and over twice the speed of FP16.
arXiv:2607. 20757v1 Announce Type: new Abstract: Transformers are known to have internal continuous symmetries that leave outputs invariant, while modifying quantization.
arXiv:2607. 21446v1 Announce Type: new Abstract: Post-training quantization (PTQ) of diffusion transformers (DiTs) to W4A4 severely degrades output quality, because activations entering each linear layer contain outliers that 4-bit formats cannot represent.
arXiv:2601. 21626v2 Announce Type: replace-cross Abstract: Post Training Quantization (PTQ), a mainstream model compression technique, often leads to the paradoxical 'low error, high loss' phenomenon because it focuses solely on minimizing quantization error.
arXiv:2507. 23035v4 Announce Type: replace Abstract: Large language models (LLMs) have demonstrated impressive capabilities across a wide range of applications, but demand substantial memory and compute resources during inference.
The paper introduces TORQUE, a framework that enhances quantization by jointly optimizing which coordinates to keep at high precision before and after applying uniform random rotations, all within a fixed bit budget. By preserving large input coordinates before rotation and the largest-magnitude coordinates after rotation, TORQUE reduces quantization error and allows efficient use of offline-optimized codebooks. The authors provide an error upper bound, prove that top‑k pre‑rotation retention is optimal for each k, and demonstrate improved accuracy‑storage tradeoffs in Gaussian models and practical tasks such as nearest‑neighbor retrieval, KV‑cache compression, and activation compression.
PRQuant introduces a training‑free, low‑overhead method for low‑bit quantization of linear layers by permuting input channels that cause the largest quantization error into contiguous tail blocks and precomputing residual weight sub‑tensors. The approach eliminates the need for online gathering during inference, converting scattered residual compensation into a regular tail‑augmented GEMM and thereby reducing latency. Experiments show that PRQuant lowers down‑projection reconstruction error and outperforms standard MXFP4 and other post‑training quantization baselines on five downstream benchmarks, improving accuracy by up to 1.24 points on Qwen3‑4B‑Instruct‑2507.
SPHQuant introduces a rotation‑free spherical weight‑only quantization framework for Vision‑Language Models, decomposing 8‑dimensional weight vectors into sign, radius, and a positive unit direction. By isolating outlier magnitudes in the radius and allocating extra precision there, it mitigates accuracy loss at extreme low bit‑widths. The method also employs a compact positive‑direction codebook with angular fine‑tuning and a hardware‑friendly GEMV kernel, achieving state‑of‑the‑art performance while boosting decode throughput by 30.3% on RTX A6000 compared to QTIP.
arXiv:2608. 06291v1 Announce Type: cross Abstract: We accelerate a family of algorithms for neural network quantization whose geometry is informed by any Kronecker-factored approximation of the Hessian.
Recent foundation models are moving toward native multimodal Vision-Language Models (VLMs), making VLMs a central form of next-generation foundation models. However, their large language backbones mak...
arXiv:2607. 18745v1 Announce Type: new Abstract: We study low-precision computation of C=AB with both factors quantized.