4 papers
Scalable Processing-Near-Memory for 1M-Token LLM Inference: CXL-Enabled KV-Cache Management Beyond GPU Limits
Dowon Kim, MinJae Lee, Janghyeon Kim +8
The expansion of context windows in large language models (LLMs) to multi-million tokens introduces severe memory and compute bottlenecks, particularly in managing the growing Key-…
Enhancing Generalization in Data-free Quantization via Mixup-class Prompting
Jiwoong Park, Chaeun Lee, Yongseok Choi +3
Post-training quantization (PTQ) improves efficiency but struggles with limited calibration data, especially under privacy constraints. Data-free quantization (DFQ) mitigates this…
PIMphony: Overcoming Bandwidth and Capacity Inefficiency in PIM-based Long-Context LLM Inference System
Hyucksung Kwon, Kyungmo Koo, Janghyeon Kim +19
The expansion of long-context Large Language Models (LLMs) creates significant memory system challenges. While Processing-in-Memory (PIM) is a promising accelerator, we identify th…
AMXFP4: Taming Activation Outliers with Asymmetric Microscaling Floating-Point for 4-bit LLM Inference
Janghwan Lee, Jiwoong Park, Jinseok Kim +4
As large language models (LLMs) grow in parameter size and context length, computation precision has been reduced from 16-bit to 4-bit to improve inference efficiency. However, thi…