We introduce HyperCLOVA X, a family of large language models (LLMs) tailored to the Korean language and culture, along with competitive capabilities in English, math, and coding. HyperCLOVA X was trained on a balanced mix of Korean, English, and code data, followed by instruction-tuning with high-quality human-annotated datasets while abiding by strict safety guidelines reflecting our commitment to responsible AI. The model is evaluated across various benchmarks, including comprehensive reasoning, knowledge, commonsense, factuality, coding, math, chatting, instruction-following, and harmlessness, in both Korean and English. HyperCLOVA X exhibits strong reasoning capabilities in Korean backed by a deep understanding of the language and cultural nuances. Further analysis of the inherent bilingual nature and its extension to multilingualism highlights the model's cross-lingual proficiency and strong generalization ability to untargeted languages, including machine translation between several language pairs and cross-lingual inference tasks. We believe that HyperCLOVA X can provide helpful guidance for regions or countries in developing their sovereign LLMs.

Large Language Models (LLMs) have recently demonstrated a remarkable success across various tasks. However, efficiently serving LLMs has been a challenge due to its large memory bottleneck, specifically in small batch inference settings (e.g. mobile devices). Weight-only quantization can be a promising approach, but sub-4 bit quantization remains a challenge due to large-magnitude activation outliers. To mitigate the undesirable outlier effect, we first propose per-IC quantization, a simple yet effective method that creates quantization groups within each input channel (IC) rather than the conventional per-output channel (OC). Our method is motivated by the observation that activation outliers affect the input dimension of the weight matrix, so similarly grouping the weights in the IC direction can isolate outliers to be within a group. We also find that activation outliers do not dictate quantization difficulty, and inherent weight sensitivities also exist. With per-IC quantization as a new outlier-friendly scheme, we then propose Adaptive Dimensions (AdaDim), a versatile quantization framework that can adapt to various weight sensitivity patterns. We demonstrate the effectiveness of AdaDim by augmenting prior methods such as Round-To-Nearest and GPTQ, showing significant improvements across various language modeling benchmarks for both base (up to +4.7% on MMLU) and instruction-tuned (up to +10% on HumanEval) LLMs.

The recent advancements in self-supervised learning, combined with the Transformer architecture, have enabled natural language processing (NLP) to achieve remarkably low perplexity. However, powerful NLP models necessitate increasing model size, leading to substantial computational and memory requirements. In this paper, we introduce an efficient inference framework tailored for large-scale generative language models. To reduce the model size, we employ a weight-only quantization strategy while preserving full precision for activations. As a result, we attain sub-4-bit quantization for each weight through non-uniform or uniform quantization techniques. Our proposed kernel, called LUT-GEMM, then accelerates quantized matrix multiplications, offering a flexible balance between compression ratio and accuracy. Unlike earlier matrix multiplication kernels that accommodated weight-only quantization, LUT-GEMM efficiently eliminates the resource-demanding dequantization process for both uniform and non-uniform quantization methods. By reducing the latency of individual GPUs and the overall inference process for large-scale language models, LUT-GEMM provides significant performance improvements in inference. The impact of LUT-GEMM is facilitated by implementing high compression ratios through low-bit quantization and efficient LUT-based operations, which decreases the number of required GPUs. For the OPT-175B model with 3-bit quantization, we show that LUT-GEMM accelerates the latency for generating each token by 2.1x compared to OPTQ, which requires costly dequantization. Consequently, LUT-GEMM enables inference of the OPT-175B model on a single GPU without noticeable degradation in accuracy or performance, while the non-quantized OPT-175B model requires a minimum of 8 GPUs.

Uniform-precision neural network quantization has gained popularity since it simplifies densely packed arithmetic unit for high computing capability. However, it ignores heterogeneous sensitivity to the impact of quantization errors across the layers, resulting in sub-optimal inference accuracy. This work proposes a novel neural architecture search called neural channel expansion that adjusts the network structure to alleviate accuracy degradation from ultra-low uniform-precision quantization. The proposed method selectively expands channels for the quantization sensitive layers while satisfying hardware constraints (e.g., FLOPs, PARAMs). Based on in-depth analysis and experiments, we demonstrate that the proposed method can adapt several popular networks channels to achieve superior 2-bit quantization accuracy on CIFAR10 and ImageNet. In particular, we achieve the best-to-date Top-1/Top-5 accuracy for 2-bit ResNet50 with smaller FLOPs and the parameter size.