Janghwan Lee
I am a Machine Learning Engineer at Qualcomm AI Research, where I work on on-device AI agents. My research interests lie broadly in Efficient AI, spanning model quantization, reduced-precision numerical formats, and hardware-aware algorithm design for scalable deep learning inference. I received my Ph.D. in Electronic Engineering from Hanyang University, where I was advised by Prof. Jungwook Choi in the Artificial Intelligence Hardware and Algorithm Lab.
Experience
Education
Publications /
⋆ denotes equal contribution. Boldface indicates the author.
- Sihwa Lee⋆, Janghwan Lee⋆, Donghoon Yoo, Jae Gon Kim, Hanyul Ryu, Soojung Ryu, and Jungwook Choi. “ReSET: Accurate Latency-Critical NVFP4 Reasoning via Step-Aware Temperature Scaling.” Conference on Neural Information Processing Systems (NeurIPS), 2026.
Large reasoning models (LRMs) improve complex problem-solving by generating long intermediate reasoning traces, but this substantially increases inference costs. NVFP4 inference offers a promising approach to reduce both computational and memory costs through hardware-supported low-precision execution. However, directly applying NVFP4 to LRMs introduces two practical limitations: reasoning accuracy degrades under quantization, and existing NVFP4 kernels do not fully realize latency benefits in small-batch autoregressive decoding. In this work, we analyze the effect of NVFP4 quantization on token-level uncertainty during reasoning. We show that quantization increases incorrect sampling at low-entropy symbolic tokens, while causing over-concentration on a small set of tokens in high-uncertainty reasoning steps. Based on this observation, we propose ReSET, a reasoning-step entropy-based temperature-scaling method that estimates step-level uncertainty online and adapts the decoding temperature using both token-level and step-level entropy signals. To address the latency gap, we further design a CUDA-core small-M NVFP4 kernel for latency-critical autoregressive decoding. Across reasoning benchmarks and model scales, ReSET improves NVFP4 reasoning accuracy by up to ~2 points over the NVFP4 baseline. Our CUDA-core small-M kernel further improves latency-critical decoding, delivering up to 2.5x kernel-level speedup over NVFP4 vLLM and approximately 2x end-to-end decoding speedup over BF16.
- Janghwan Lee, Sihwa Lee, Jinseok Kim, Yongjik Kim, Jieun Lim, Jinwook Oh, and Jungwook Choi. “ReQAT: Achieving Full-Precision Reasoning Accuracy with 4-bit Floating-Point Quantization-Aware Training.” International Conference on Machine Learning (ICML), 2026. (Spotlight & Oral)
Large Reasoning Models (LRMs) achieve strong problem-solving through long chain-of-thought, but their deployment is constrained by the high cost of full-precision inference and growing KV cache footprints. Microscaled FP4 formats enable efficient FP4 deployment; however, fully quantizing weights, activations, and KV caches (W4A4KV4) causes severe reasoning degradation that existing PTQ and QAT fail to recover. We identify that FP4 failures concentrate on low-entropy tokens—precise symbolic commitments such as digits and operators—where quantization noise inflates sampling errors that cascade through reasoning traces. Based on this insight, we propose ReQAT, a reasoning-centric FP4 training framework with three components: (i) Trace-Aligned QAT (TAQ), which revisits identical reasoning traces to focus updates on critical low-entropy decisions; (ii) Selective Entropy Minimization (SEM), which reinforces confidence at low-entropy positions; and (iii) Q-FIT, a quantization-friendly initialization that jointly calibrates RoPE-consistent KV cache transformations to stabilize QAT. Under the same training budget, ReQAT not only recovers but surpasses BF16 fine-tuning accuracy—achieving while delivering up to 3.9x throughput speedup on NVIDIA DGX Spark and 3.1x on B200.
- Janghwan Lee, Jiwoong Park, Jinseok Kim, Yongjik Kim, Jungju Oh, Jinwook Oh, and Jungwook Choi. “AMXFP4: Taming Activation Outliers with Asymmetric Microscaling Floating-Point for 4-bit LLM Inference.” Findings of the Association for Computational Linguistics (ACL Findings), 2025.
Scaling Large Language Models (LLMs) with extended context lengths has increased the need for efficient low-bit quantization to manage their substantial computational demands. However, reducing precision to 4 bits frequently degrades performance due to activation outliers. To address this, we propose Asymmetric Microscaling 4-bit Floating-Point (AMXFP4) for efficient LLM inference. This novel data format leverages asymmetric shared scales to mitigate outliers while naturally capturing the asymmetry introduced by group-wise quantization. Unlike conventional 4-bit quantization methods that rely on data rotation and costly calibration, AMXFP4 uses asymmetric shared scales for direct 4-bit casting, achieving near-ideal quantization accuracy across various LLM tasks, including multi-turn conversations, long-context reasoning, and visual question answering.
- Geonho Lee⋆, Janghwan Lee⋆, Sukjin Hong⋆, Minsoo Kim, Euijai Ahn, Du-Seong Chang, and Jungwook Choi. “RILQ: Rank-Insensitive LoRA-based Quantization Error Compensation for Boosting 2-bit Large Language Model Accuracy.” AAAI Conference on Artificial Intelligence (AAAI), 2025.
Low-rank adaptation (LoRA) has become the dominant method for parameter-efficient LLM fine-tuning, with LoRA-based quantization error compensation (LQEC) emerging as a powerful tool for recovering accuracy in compressed LLMs. However, LQEC has underperformed in sub-4-bit scenarios, with no prior investigation into understanding this limitation. We propose RILQ (Rank-Insensitive LoRA-based Quantization Error Compensation) to understand fundamental limitation and boost 2-bit LLM accuracy. Based on rank analysis revealing model-wise activation discrepancy loss's rank-insensitive nature, RILQ employs this loss to adjust adapters cooperatively across layers, enabling robust error compensation with low-rank adapters. Evaluations on LLaMA-2 and LLaMA-3 demonstrate RILQ's consistent improvements in 2-bit quantized inference across various state-of-the-art quantizers and enhanced accuracy in task-specific fine-tuning. RILQ maintains computational efficiency comparable to existing LoRA methods, enabling adapter-merged weight-quantized LLM inference with significantly enhanced accuracy, making it a promising approach for boosting 2-bit LLM performance.
- Janghwan Lee⋆, Seongmin Park⋆, Sukjin Hong, Minsoo Kim, Du-Seong Chang, and Jungwook Choi. “Improving Conversational Abilities of Quantized Large Language Models via Direct Preference Alignment.” Annual Meeting of the Association for Computational Linguistics (ACL), 2024. (Oral)
The rapid advancement of large language models (LLMs) has facilitated their transformation into conversational chatbots that can grasp contextual nuances and generate pertinent sentences, closely mirroring human values through advanced techniques such as instruction tuning and reinforcement learning from human feedback (RLHF). However, the computational efficiency required for LLMs, achieved through techniques like post-training quantization (PTQ), presents challenges such as token-flipping that can impair chatbot performance. In response, we propose a novel preference alignment approach, quantization-aware direct preference optimization (QDPO), that aligns quantized LLMs with their full-precision counterparts, improving conversational abilities. Evaluated on two instruction-tuned LLMs in various languages, QDPO demonstrated superior performance in improving conversational abilities compared to established PTQ and knowledge-distillation fine-tuning techniques, marking a significant step forward in the development of efficient and effective conversational LLMs.
- Dong-eon Won⋆, Yeeun Kim⋆, Janghwan Lee, Minjae Lee, Jonghyun Bae, Jongjoo Park, Jeongyong Song, and Jungwook Choi. “ISP2DLA: Automated Deep Learning Accelerator Design for On-Sensor Image Signal Processing.” IEEE International Conference on Application-Specific Systems, Architectures and Processors (ASAP), 2024.
Deep neural network-based image signal processing (ISP-DNN) improves image quality with techniques such as demosaicing, but these models pose substantial computational and memory challenges when implemented on CMOS image sensors, particularly due to the high-resolution inputs that increase memory requirements for activations. Layer fusion reduces memory usage by combining consecutive processing steps, yet it increases computational demands, a critical issue in resource-limited on-sensor environments. To address these challenges, we introduce ISP2DLA, an automated deep learning accelerator design framework that balances computational and memory demands for on-sensor ISP. This framework optimizes hardware designs by adjusting line buffer sizes and the number of MAC units, reducing gate counts by 14-79% across two ISP-DNN models, thus enabling efficient on-sensor ISP model inference within constrained resources.
- Minjae Lee, Seongmin Park, Hyungmin Kim, Minyong Yoon, Janghwan Lee, Junwon Choi, Nam Sung Kim, Mingu Kang, and Jungwook Choi. “SPADE: Sparse Pillar-based 3D Object Detection Accelerator for Autonomous Driving.” IEEE International Symposium on High-Performance Computer Architecture (HPCA), 2024.
3D object detection using point cloud (PC) data is essential for perception pipelines of autonomous driving, where efficient encoding is key to meeting stringent resource and latency requirements. PointPillars, a widely adopted bird's-eye view (BEV) encoding, aggregates 3D point cloud data into 2D pillars for fast and accurate 3D object detection. However, the state-of-the-art methods employing PointPillars overlook the inherent sparsity of pillar encoding where only a valid pillar is encoded with a vector of channel elements, missing opportunities for significant computational reduction. Meanwhile, current sparse convolution accelerators are designed to handle only element-wise activation sparsity and do not effectively address the vector sparsity imposed by pillar encoding. In this paper, we propose SPADE, an algorithm-hardware co-design strategy to maximize vector sparsity in pillar-based 3D object detection and accelerate vector-sparse convolution commensurate with the improved sparsity. SPADE consists of three components: (1) a dynamic vector pruning algorithm balancing accuracy and computation savings from vector sparsity, (2) a sparse coordinate management hardware transforming 2D systolic array into a vector-sparse convolution accelerator, and (3) sparsity-aware dataflow optimization tailoring sparse convolution schedules for hardware efficiency. Taped-out with a commercial technology, SPADE saves the amount of computation by 36.3-89.2% for representative 3D object detection networks and benchmarks, leading to 1.3-10.9x speedup and 1.5-12.6x energy savings compared to the ideal dense accelerator design. These sparsity-proportional performance gains equate to 4.1-28.8x speedup and 90.2-372.3x energy savings compared to the counterpart server and edge platforms.
- Youngdeok Hwang⋆, Janghwan Lee⋆, Jiwoong Park, Jieun Lim, and Jungwook Choi. “Searching Optimal Floating-Point Format for Sub-8-Bit Large Language Model Inference.” International Conference on Electronics, Information, and Communication (ICEIC), 2024. (Oral)
Large Language Models (LLMs) have shown remarkable success in various natural language processing tasks. However, their extensive parameter count leads to significant memory and computational demands. To tackle these challenges, there is growing interest in employing post-training quantization (PTQ) with reduced-precision floating-point (FP) operations. Yet, the optimal FP configuration remains a topic of debate. Existing studies often overlook a thorough analysis of the diverse data distributions found in LLMs and the crucial design choice, denormal. In this paper, we conduct a comprehensive examination of the various data distributions within LLMs and the significance of denormal representation, presenting a mixed-format floating-point framework. Our proposed framework allows for sub-8-bit inference with minimal performance degradation in language modeling and reasoning tasks across a broad spectrum of LLMs.
- Janghwan Lee⋆, Minsoo Kim⋆, Seungcheol Baek, Seok Joong Hwang, Wonyong Sung, and Jungwook Choi. “Enhancing Computation Efficiency in Large Language Models through Weight and Activation Quantization.” Conference on Empirical Methods in Natural Language Processing (EMNLP), 2023.
Large Language Models (LLMs) are proficient in natural language processing tasks, but their deployment is often restricted by extensive parameter sizes and computational demands. This paper focuses on post-training quantization (PTQ) in LLMs, specifically 4-bit weight and 8-bit activation (W4A8) quantization, to enhance computational efficiency—a topic less explored compared to weight-only quantization. We present two innovative techniques: activation-quantization-aware scaling (AQAS) and sequence-length-aware calibration (SLAC) to enhance PTQ by considering the combined effects on weights and activations and aligning calibration sequence lengths to target tasks. Moreover, we introduce dINT, a hybrid data format combining integer and denormal representations, to address the underflow issue in W4A8 quantization, where small values are rounded to zero. Through rigorous evaluations of LLMs, including OPT and LLaMA, we demonstrate that our techniques significantly boost task accuracies to levels comparable with full-precision models. By developing arithmetic units compatible with dINT, we further confirm that our methods yield a 2x hardware efficiency improvement compared to 8-bit integer MAC unit.
- Minsoo Kim, Sihwa Lee, Janghwan Lee, Sukjin Hong, Du-Seong Chang, Wonyong Sung, and Jungwook Choi. “Token-Scaled Logit Distillation for Ternary Weight Generative Language Models.” Conference on Neural Information Processing Systems (NeurIPS), 2023.
Generative Language Models (GLMs) have shown impressive performance in tasks such as text generation, understanding, and reasoning. However, the large model size poses challenges for practical deployment. To solve this problem, Quantization-Aware Training (QAT) has become increasingly popular. However, current QAT methods for generative models have resulted in a noticeable loss of accuracy. To counteract this issue, we propose a novel knowledge distillation method specifically designed for GLMs. Our method, called token-scaled logit distillation, prevents overfitting and provides superior learning from the teacher model and ground truth. This research marks the first evaluation of ternary weight quantization-aware training of large-scale GLMs with less than 1.0 degradation in perplexity and achieves enhanced accuracy in tasks like common-sense QA and arithmetic reasoning as well as natural language understanding.
- Janghwan Lee, Youngdeok Hwang, and Jungwook Choi. “Finding Optimal Numerical Format for Sub-8-bit Post-Training Quantization of Vision Transformers.” IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023.
Vision Transformers (ViTs) have gained significant attention for their exceptional model accuracies on computer vision applications, but their demanding memory requirements and computational complexity have hindered active deployment. Post-training quantization (PTQ) is a practical method to tackle this challenge by directly reducing ViT's bit-precision. However, diverse data characteristics across different operations of ViT cannot be well captured solely by a single numerical format (fixed or floating-point). This work proposes an analytical framework that optimizes the numerical format of each matrix multiplication of ViTs for mixed-format sub-8bit quantization. The extensive evaluation demonstrates that the proposed method can reduce the PTQ error and achieve state-of-the-art accuracy for popular ViT models.
- Janghyeon Kim, Janghwan Lee, JeongHo Han, Sangheon Lee, and Jungwook Choi. “Range-Invariant Approximation of Non-Linear Operations for Efficiently Fine-tuning BERT.” ACM/IEEE Design Automation Conference (DAC), 2023.
This paper proposes a range-invariant approximation of non-linear operations for training computations of Transformer-based large language models. The proposed method decomposes the approximation into the scaling and the range-invariant resolution for LUT approximation, covering diverse data ranges of non-linear operations with drastically reduced LUT entries during task-dependent BERT fine-tuning. We demonstrate that the proposed method robustly approximates all the non-linear operations of BERT without score degradation on challenging GLUE benchmarks using only a single-entry LUT, facilitating 52% area savings in hardware implementation.
- Janghwan Lee and Jungwook Choi. “Optimizing Exponent Bias for Sub-8bit Floating-Point Inference of Fine-tuned Transformers.” IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS), 2022. (Oral)
The Transformer-based fine-tuned neural networks have demonstrated remarkable success in natural language processing (NLP) at the cost of a substantial computational burden. Post-training quantization (PTQ) is a promising technique to reduce the computational cost without expensive re-training. But prior works either demand complex calibration or suffer noticeable accuracy degradation. This paper proposes a practical method for sub-8bit floating-point (FP) PTQ. The proposed method optimizes the exponent bias to minimize quantization error in terms of signal-to-quantization noise ratio (SQNR) progressively like stochastic gradient descent. We evaluate that the proposed method achieves close to full-precision model accuracy for 6 to 8 bit FP PTQ of fine-tuned BERT on GLUE and SQuAD tasks with negligible run-time overhead.
Honors & Awards
- Gold Reviewer, International Conference on Machine Learning (ICML)2026
- Excellence in Research and Education Award, ENRICH IT Award, 4th Stage BK212025
- First Place, Model Compression Track, AI Grand Challenge, Ministry of Science and ICT, Korea2020
- Integrated Ph.D. Course Scholarship (Full Tuition), Hanyang University2020–2022
- Research Scholarship (KRW 24M total), ISRC2020–2022
Technical Skills
- Languages:
- Python, C, C++
- Frameworks:
- PyTorch, Hugging Face Transformers
- Platforms:
- Linux, Git, Docker