Ik Joon Chang

Dongyoung Lee, Seungkyu Choi, Ik Joon Chang

Large-scale language models (LLMs) excel in language processing tasks but face deployment challenges due to high memory and computational demands. While low-bit quantization, such as 4-bit techniques, offers a potential solution, these methods often suffer from significant accuracy loss or require considerable effort for implementation such as reordering, rotation, etc. To address these challenges, we propose QRazor, a simple yet effective quantization scheme that enables 4-bit quantization of weights, activations, and KV cache in transformer-based LLMs. QRazor operates in two stages: first, quantizing data using 8 or 16-bit integers as a basis with absolute max scaling to preserve accuracy close to full-precision models, and second, compressing the quantized data to 4-bit using our significant data razoring (SDR) technique, which retains only the four most salient bits. Without any additional requirment of fine-tuning or additional training, QRazor achieves performance similar or better compared to state-of-the-art in 4-bit quantization method, surpassing Smoothquant and QLLM by over 12 points and Quarot(RTN) by more than 2.9 points in zero-shot reasoning task accuracy on the LLaMA2-7B model. Additionally, we introduce an integer-based arithmetic unit optimized for QRazor, allowing direct low-precision operations on SDR data without decompression.

Donghoon Kim, Dongyoung Lee, Ik Joon Chang et al.

Diffusion models achieve high-quality image generation but face deployment challenges due to their high computational requirements. Although 8-bit outlier-aware post-training quantization (PTQ) matches full-precision performance, extending PTQ to 4 bits remains challenging. Larger step sizes in 4-bit quantization amplify rounding errors in dense, low-magnitude activations, leading to the loss of fine-grained textures. We hypothesize that not only outliers but also small activations are critical for texture fidelity. To this end, we propose Quantization via Residual Truncation and Zero Suppression (QuaRTZ), a 4-bit PTQ scheme for diffusion models. QuaRTZ applies 8-bit min-max quantization for outlier handling and compresses to 4 bits via leading-zero suppression to retain LSBs, thereby preserving texture details. Our approach reduces rounding errors and improves quantization efficiency by balancing outlier preservation and LSB precision. Both theoretical derivations and empirical evaluations demonstrate the generalizability of QuaRTZ across diverse activation distributions. Notably, 4-bit QuaRTZ achieves an FID of 6.98 on FLUX.1-schnell, outperforming SVDQuant that requires auxiliary FP16 branches.

Donghyeon Yi, Seoyoung Lee, Jongho Kim et al.

Encoder-based transformers, powered by self-attention layers, have revolutionized machine learning with their context-aware representations. However, their quadratic growth in computational and memory demands presents significant bottlenecks. Analog-Mixed-Signal Process-in-Memory (AMS-PiM) architectures address these challenges by enabling efficient on-chip processing. Traditionally, AMS-PiM relies on Quantization-Aware Training (QAT), which is hardware-efficient but requires extensive retraining to adapt models to AMS-PiMs, making it increasingly impractical for transformer models. Post-Training Quantization (PTQ) mitigates this training overhead but introduces significant hardware inefficiencies. PTQ relies on dequantization-quantization (DQ-Q) processes, floating-point units (FPUs), and high-ENOB (Effective Number of Bits) analog-to-digital converters (ADCs). Particularly, High-ENOB ADCs scale exponentially in area and energy ($2^{ENOB}$), reduce sensing margins, and increase susceptibility to process, voltage, and temperature (PVT) variations, further compounding PTQ's challenges in AMS-PiM systems. To overcome these limitations, we propose RAP, an AMS-PiM architecture that eliminates DQ-Q processes, introduces FPU- and division-free nonlinear processing, and employs a low-ENOB-ADC-based sparse Matrix Vector multiplication technique. Using the proposed techniques, RAP improves error resiliency, area/energy efficiency, and computational speed while preserving numerical stability. Experimental results demonstrate that RAP outperforms state-of-the-art GPUs and conventional PiM architectures in energy efficiency, latency, and accuracy, making it a scalable solution for the efficient deployment of transformers.