GPTVQ: The Blessing of Dimensionality for LLM Quantization
read the original abstract
In this work we show that the size versus accuracy trade-off of neural network quantization can be significantly improved by increasing the quantization dimensionality. We propose the GPTVQ method, a new fast method for post-training vector quantization (VQ) that scales well to Large Language Models (LLMs). Our method interleaves quantization of one or more columns with updates to the remaining unquantized weights, using information from the Hessian of the per-layer output reconstruction MSE. Quantization codebooks are initialized using an efficient data-aware version of the EM algorithm. The codebooks are then updated, and further compressed by using integer quantization and SVD-based compression. GPTVQ establishes a new state-of-the art in the size vs accuracy trade-offs on a wide range of LLMs such as Llama-v2 and Mistral. Furthermore, our method is efficient: on a single H100 it takes between 3 and 11 hours to process a Llamav2-70B model, depending on quantization setting. Lastly, with on-device timings for VQ decompression on a mobile CPU we show that VQ leads to improved latency compared to using a 4-bit integer format.
This paper has not been read by Pith yet.
Forward citations
Cited by 10 Pith papers
-
Statistically-Lossless Quantization of Large Language Models
SLQ achieves task-lossless LLM quantization below 4 bits per parameter and distribution-lossless at 5-6 bits on average, with 1.7-3.6x speedups over FP16.
-
LBLLM: Lightweight Binarization of Large Language Models via Three-Stage Distillation
LBLLM achieves better accuracy than prior binarization methods for LLMs by decoupling weight and activation quantization through initialization, layer-wise distillation, and learnable activation scaling.
-
Initialisation Determines the Basin: Efficient Codebook Optimisation for Extreme LLM Quantization
Output-aware EM initialization for codebooks in additive quantization avoids poor optimization basins and yields better 2-bit compressed LLMs across Llama and Qwen models.
-
CoreQ: Learning-Free Mismatch Correction and Successive Rounding for Quantization
CoreQ delivers adaptive mismatch correction via closed-form geometric coefficient and successive rounding to improve PTQ accuracy for large language models.
-
OmniZip: Audio-Guided Dynamic Token Compression for Fast Omnimodal Large Language Models
OmniZip introduces an audio-guided dynamic token compression framework that achieves 3.42X inference speedup and 1.4X memory reduction for omnimodal LLMs without any training.
-
BTC-LLM: Efficient Sub-1-Bit LLM Quantization via Learnable Transformation and Binary Codebook
BTC-LLM uses a binary codebook for pattern clustering and a learnable transformation to achieve 0.7-1.11 bit LLM quantization while limiting accuracy loss to a few percent on LLaMA and Qwen models.
-
OffQ: Taming Structured Outliers in LLM Quantization by Offsetting
OffQ mitigates structured activation outliers in LLMs via PCA-based rotation and shared offset absorption to support effective W4A4KV4 uniform quantization.
-
EVA: Accelerating LLM Decoding via an Efficient Vector Quantization Architecture
EVA is a vector-quantization hardware architecture that transforms LLM decoding from GEMV to GEMM via direct codebook dot products and conflict-free output buffering, claiming up to 11.17x speedup over prior lookup designs.
-
MGVQ: Synergizing Multi-dimensional Sensitivity-Aware and Gradient-Hessian Fusion for Vector Quantization
MGVQ introduces sensitivity-aware structured mixed-precision VQ and gradient-aware second-order error compensation using Kronecker and Block-LDL decompositions, reporting up to 4.9 point gains over prior methods at 2-...
-
31.1 A 14.08-to-135.69Token/s ReRAM-on-Logic Stacked Outlier-Free Large-Language-Model Accelerator with Block-Clustered Weight-Compression and Adaptive Parallel-Speculative-Decoding
A ReRAM-on-logic stacked chip delivers 14.08-135.69 tokens/s LLM inference with block-clustered compression and adaptive parallel speculative decoding, yielding 4.46-7.17x speedup over standard methods.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.