arXiv subjects

Sanghoon Kim

Publications and source records attributed to Sanghoon Kim.

At least 19 recordsLinked to original sources

arXiv2026

Solar Open 2 Technical Report

We present Solar Open 2, a 250B-A15B Mixture-of-Experts language model built for long-horizon agentic tasks, scaled up from Solar Open 1 (Solar Open 100B). To hold entire agent trajectories in a single context, Solar Open 2 reaches a 1M-token window through a hybrid attention stack that interleaves one softmax layer among every three linear-attention layers, using no positional encoding and a gated delta rule extended to negative eigenvalues. To train at this scale under a fixed compute budget, we make training efficient in two ways: a stronger starting point, and higher-value data. For the starting point, we initialize Solar Open 2 from Solar Open 1, transferring the 5.69B-parameter shared skeleton that survives the architectural change and learning everything else through full pre-training. For the data, we curate for value per token: quality- and rarity-aware data curation and mixture-ratio optimization refine a 20T pool into a 10T mixture that, at equal token budget, outperforms the Solar Open 1 recipe. To build its agent skills, we train twelve domain specialists across purpose-built scenarios, then consolidate them into a single model by Multi-teacher On-Policy Distillation (MOPD). Against comparably sized open-weight models on English benchmarks, Solar Open 2 leads on MMLU-Pro, LiveCodeBench, and the APEX-Agents agentic suite, and stays competitive with the strongest (DeepSeek-V4-Flash and MiMo-V2.5) elsewhere. On Korean benchmarks, Solar Open 2 records the highest average of any model compared, including fast-tier closed APIs, and on Ko-GDPval, an in-house Korean officework-agent benchmark, it is competitive with DeepSeek-V4-Pro (1.6T) at less than a sixth of its size.

cs.CL↗

arXiv2026

Solar Open Technical Report

We introduce Solar Open, a 102B-parameter bilingual Mixture-of-Experts language model for underserved languages. Solar Open demonstrates a systematic methodology for building competitive LLMs by addressing three interconnected challenges. First, to train effectively despite data scarcity for underserved languages, we synthesize 4.5T tokens of high-quality, domain-specific, and RL-oriented data. Second, we coordinate this data through a progressive curriculum jointly optimizing composition, quality thresholds, and domain coverage across 20 trillion tokens. Third, to enable reasoning capabilities through scalable RL, we apply our proposed framework SnapPO for efficient optimization. Across benchmarks in English and Korean, Solar Open achieves competitive performance, demonstrating the effectiveness of this methodology for underserved language AI development.

cs.CL↗

arXiv2025

BemaGANv2: Discriminator Combination Strategies for GAN-based Vocoders in Long-Term Audio Generation

This paper presents BemaGANv2, an advanced GAN-based vocoder designed for high-fidelity and long-term audio generation, with a focus on systematic evaluation of discriminator combination strategies. Long-term audio generation is critical for applications in Text-to-Music (TTM) and Text-to-Audio (TTA) systems, where maintaining temporal co- herence, prosodic consistency, and harmonic structure over extended durations remains a significant challenge. Built upon the original BemaGAN architecture, BemaGANv2 incorporates major architectural innovations by replacing traditional ResBlocks in the generator with the Anti-aliased Multi-Periodicity composition (AMP) module, which internally applies the Snake activation function to better model periodic structures. In the discriminator framework, we integrate the Multi-Envelope Discriminator (MED), a novel architecture we proposed, to extract rich temporal en- velope features crucial for periodicity detection. Coupled with the Multi-Resolution Discriminator (MRD), this com- bination enables more accurate modeling of long-range dependencies in audio. We systematically evaluate various discriminator configurations, including Multi-Scale Discriminator (MSD) + MED, MSD + MRD, and Multi-Period Discriminator (MPD) + MED + MRD, using objective metrics (Fr\'echet Audio Distance (FAD), Structural Similar- ity Index (SSIM), Pearson Correlation Coefficient (PCC), Mel-Cepstral Distortion (MCD), Multi-Resolution STFT (M-STFT), Periodicity error (Periodicity)) and subjective evaluations (MOS, SMOS). To support reproducibility, we provide detailed architectural descriptions, training configurations, and complete implementation details. The code, pre-trained models, and audio demo samples are available at: https://github.com/dinhoitt/BemaGANv2.

cs.SD↗

arXiv2024

Open Ko-LLM Leaderboard: Evaluating Large Language Models in Korean with Ko-H5 Benchmark

This paper introduces the Open Ko-LLM Leaderboard and the Ko-H5 Benchmark as vital tools for evaluating Large Language Models (LLMs) in Korean. Incorporating private test sets while mirroring the English Open LLM Leaderboard, we establish a robust evaluation framework that has been well integrated in the Korean LLM community. We perform data leakage analysis that shows the benefit of private test sets along with a correlation study within the Ko-H5 benchmark and temporal analyses of the Ko-H5 score. Moreover, we present empirical support for the need to expand beyond set benchmarks. We hope the Open Ko-LLM Leaderboard sets precedent for expanding LLM evaluation to foster more linguistic diversity.

cs.CL↗

arXiv2024

SOLAR 10.7B: Scaling Large Language Models with Simple yet Effective Depth Up-Scaling

We introduce SOLAR 10.7B, a large language model (LLM) with 10.7 billion parameters, demonstrating superior performance in various natural language processing (NLP) tasks. Inspired by recent efforts to efficiently up-scale LLMs, we present a method for scaling LLMs called depth up-scaling (DUS), which encompasses depthwise scaling and continued pretraining. In contrast to other LLM up-scaling methods that use mixture-of-experts, DUS does not require complex changes to train and inference efficiently. We show experimentally that DUS is simple yet effective in scaling up high-performance LLMs from small ones. Building on the DUS model, we additionally present SOLAR 10.7B-Instruct, a variant fine-tuned for instruction-following capabilities, surpassing Mixtral-8x7B-Instruct. SOLAR 10.7B is publicly available under the Apache 2.0 license, promoting broad access and application in the LLM field.

cs.CL↗

arXiv2024

sDPO: Don't Use Your Data All at Once

As development of large language models (LLM) progresses, aligning them with human preferences has become increasingly important. We propose stepwise DPO (sDPO), an extension of the recently popularized direct preference optimization (DPO) for alignment tuning. This approach involves dividing the available preference datasets and utilizing them in a stepwise manner, rather than employing it all at once. We demonstrate that this method facilitates the use of more precisely aligned reference models within the DPO training framework. Furthermore, sDPO trains the final model to be more performant, even outperforming other popular LLMs with more parameters.

cs.CL↗

arXiv2022

Efficient conversion of orbital Hall current to spin current for spin-orbit torque switching

Spin Hall effect, an electric generation of spin current, allows for efficient control of magnetization. Recent theory revealed that orbital Hall effect creates orbital current, which can be much larger than spin Hall-induced spin current. However, orbital current cannot directly exert a torque on a ferromagnet, requiring a conversion process from orbital current to spin current. Here, we report two effective methods of the conversion through spin-orbit coupling engineering, which allows us to unambiguously demonstrate orbital-current-induced spin torque, or orbital Hall torque. We find that orbital Hall torque is greatly enhanced by introducing either a rare-earth ferromagnet Gd or a Pt interfacial layer with strong spin-orbit coupling in Cr/ferromagnet structures, indicating that the orbital current generated in Cr is efficiently converted into spin current in the Gd or Pt layer. Furthermore, we show that the orbital Hall torque can facilitate the reduction of switching current of perpendicular magnetization in spin-orbit-torque-based spintronic devices.

cond-mat.mtrl-sci↗

arXiv2022

Interactive Multi-Class Tiny-Object Detection

Annotating tens or hundreds of tiny objects in a given image is laborious yet crucial for a multitude of Computer Vision tasks. Such imagery typically contains objects from various categories, yet the multi-class interactive annotation setting for the detection task has thus far been unexplored. To address these needs, we propose a novel interactive annotation method for multiple instances of tiny objects from multiple classes, based on a few point-based user inputs. Our approach, C3Det, relates the full image context with annotator inputs in a local and global manner via late-fusion and feature-correlation, respectively. We perform experiments on the Tiny-DOTA and LCell datasets using both two-stage and one-stage object detection architectures to verify the efficacy of our approach. Our approach outperforms existing approaches in interactive annotation, achieving higher mAP with fewer clicks. Furthermore, we validate the annotation efficiency of our approach in a user study where it is shown to be 2.85x faster and yield only 0.36x task load (NASA-TLX, lower is better) compared to manual annotation. The code is available at https://github.com/ChungYi347/Interactive-Multi-Class-Tiny-Object-Detection.

cs.CV↗

arXiv2020

A community-powered search of machine learning strategy space to find NMR property prediction models

The rise of machine learning (ML) has created an explosion in the potential strategies for using data to make scientific predictions. For physical scientists wishing to apply ML strategies to a particular domain, it can be difficult to assess in advance what strategy to adopt within a vast space of possibilities. Here we outline the results of an online community-powered effort to swarm search the space of ML strategies and develop algorithms for predicting atomic-pairwise nuclear magnetic resonance (NMR) properties in molecules. Using an open-source dataset, we worked with Kaggle to design and host a 3-month competition which received 47,800 ML model predictions from 2,700 teams in 84 countries. Within 3 weeks, the Kaggle community produced models with comparable accuracy to our best previously published "in-house" efforts. A meta-ensemble model constructed as a linear combination of the top predictions has a prediction accuracy which exceeds that of any individual model, 7-19x better than our previous state-of-the-art. The results highlight the potential of transformer architectures for predicting quantum mechanical (QM) molecular properties.

physics.comp-ph↗

arXiv2020

GUIComp: A GUI Design Assistant with Real-Time, Multi-Faceted Feedback

Users may face challenges while designing graphical user interfaces, due to a lack of relevant experience and guidance. This paper aims to investigate the issues that users with no experience face during the design process, and how to resolve them. To this end, we conducted semi-structured interviews, based on which we built a GUI prototyping assistance tool called GUIComp. This tool can be connected to GUI design software as an extension, and it provides real-time, multi-faceted feedback on a user's current design. Additionally, we conducted two user studies, in which we asked participants to create mobile GUIs with or without GUIComp, and requested online workers to assess the created GUIs. The experimental results show that GUIComp facilitated iterative design and the participants with GUIComp had better a user experience and produced more acceptable designs than those who did not.

cs.HC↗

arXiv2019

Enhanced perpendicular magnetocrystalline anisotropy energy in an artificial magnetic material with bulk spin-momentum coupling

We systematically investigate the perpendicular magnetocrystalline anisotropy (MCA) in Co$-$Pt/Pd-based multilayers. Our magnetic measurement data shows that the asymmetric Co/Pd/Pt multilayer has a significantly larger perpendicular magnetic anisotropy (PMA) energy compared to the symmetric Co/Pt and Co/Pd multilayer samples. We further support this experiment by first principles calculations on the CoPt$_2$, CoPd$_2$, and CoPtPd, which are composite bulk materials that consist of three atomic layers in a unit cell, Pt/Co/Pt, Pd/Co/Pd, Pt/Co/Pd, respectively. By estimating the contribution of bulk spin-momentum coupling to the MCA energy, we show that the CoPtPd multilayer with the symmetry breaking has a significantly larger perpendicular magnetic anisotropy (PMA) energy than the other multilayers that are otherwise similar but lack the symmetry breaking. This observation thus provides an evidence of the PMA enhancement due to the structural inversion symmetry breaking and highlights the asymmetric CoPtPd as the first artificial magnetic material with bulk spin-momentum coupling, which opens a new pathway toward the design of materials with strong PMA.

cond-mat.mtrl-sci↗

arXiv2018

Enhanced spin-orbit torque via interface engineering in Pt/CoFeB/MgO heterostructures

Spin-orbit torque facilitates efficient magnetization switching via an in-plane current in perpendicularly magnetized heavy metal/ferromagnet heterostructures. The efficiency of spin-orbit-torque-induced switching is determined by the charge-to-spin conversion arising from either bulk or interfacial spin-orbit interactions, or both. Here, we demonstrate that the spin-orbit torque and the resultant switching efficiency in Pt/CoFeB systems are significantly enhanced by an interfacial modification involving Ti insertion between the Pt and CoFeB layers. Spin pumping and X-ray magnetic circular dichroism experiments reveal that this enhancement is due to an additional interface-generated spin current of the nonmagnetic interface and/or improved spin transparency achieved by suppressing the proximity-induced moment in the Pt layer. Our results demonstrate that interface engineering affords an effective approach to improve spin-orbit torque and thereby magnetization switching efficiency.

cond-mat.mtrl-sci↗

arXiv2017

Distinct domain-wall motion between creep and flow regimes near the angular momentum compensation temperature of ferrimagnet

We investigate a magnetic domain-wall (DW) motion in two dynamic regimes, creep and flow regimes, near the angular momentum compensation temperature (T_A) of ferrimagnet. In the flow regime, the DW speed shows sharp increase at T_A due to the emergence of antiferromagnetic DW dynamics. In the creep regime, however, the DW speed exhibits a monotonic increase with increasing the temperature. This result suggests that, in the creep regime, the thermal activation process governs the DW dynamics even near T_A. Our result unambiguously shows the distinct behavior of ferrimagnetic DW motion depending on the dynamic regime, which is important for emerging ferrimagnet-based spintronic applications.

cond-mat.mtrl-sci↗

arXiv2017

Origin of threshold current density for asymmetric magnetoresistance in Pt/Py bilayers

An asymmetric magnetoresistance (MR) is investigated in Py/Pt bilayers. The asymmetric MR linearly increases with current density up to a threshold, and increases more rapidly above the threshold. To reveal the origin of threshold behavior, we investigate the magnetic field dependence of the asymmetric MR. It is found that the magnetic field strongly suppresses the asymmetric MR only above the threshold current density. Micromagnetic simulation reveals that the reduction of magnetization due to the spin-torque oscillation can be the origin of the threshold behavior of asymmetric MR.

cond-mat.mtrl-sci↗

arXiv2017

Contributions of Co and Fe orbitals to Perpendicular Magnetic Anisotropy of MgO/CoFeB Bilayers with Spin-Orbit-Torque-Related (Ta, W, IrMn, and Ti) Underlayers

We study the perpendicular magnetic anisotropy (PMA) of the CoFeB/MgO bilayers in contact with W, Ta, IrMn and Ti which has been suggested as the spin-orbit-torque-related underlayers. The saturation magnetization of the CoFeB depends on the underlayer materials due to formation of a dead-layer, affecting PMA strength of each film. The x-ray magnetic circular dichroism measurement reveals that the interfacial intermixing suppresses only the perpendicular orbital moment of Fe, while the intermixing simultaneously suppresses both the perpendicular and in-plane orbital moments of Co.

cond-mat.mtrl-sci↗

arXiv2017

Microscopic Origin of Interfacial Dzyaloshinskii-Moriya Interaction

Chiral spin textures at the interface between ferromagnetic and heavy nonmagnetic metals, such as Neel-type domain walls and skyrmions, have been studied intensively because of their great potential for future nanomagnetic devices. The Dyzaloshinskii-Moriya interaction (DMI) is an essential phenomenon for the formation of such chiral spin textures. In spite of recent theoretical progress aiming at understanding the microscopic origin of the DMI, an experimental investigation unravelling the physics at stake is still required. Here, we experimentally demonstrate the close correlation of the DMI with the anisotropy of the orbital magnetic moment and with the magnetic dipole moment of the ferromagnetic metal. The density functional theory and the tight-binding model calculations reveal that asymmetric electron occupation in orbitals gives rise to this correlation.

cond-mat.mtrl-sci↗

arXiv2017

Fast domain wall motion induced by antiferromagnetic spin dynamics at the angular momentum compensation temperature of ferrimagnets

Antiferromagnetic spintronics is an emerging research field which aims to utilize antiferromagnets as core elements in spintronic devices. A central motivation toward this direction is that antiferromagnetic spin dynamics is expected to be much faster than ferromagnetic counterpart because antiferromagnets have higher resonance frequencies than ferromagnets. Recent theories indeed predicted faster dynamics of antiferromagnetic domain walls (DWs) than ferromagnetic DWs. However, experimental investigations of antiferromagnetic spin dynamics have remained unexplored mainly because of the immunity of antiferromagnets to magnetic fields. Furthermore, this immunity makes field-driven antiferromagnetic DW motion impossible despite rich physics of field-driven DW dynamics as proven in ferromagnetic DW studies. Here we show that fast field-driven antiferromagnetic spin dynamics is realized in ferrimagnets at the angular momentum compensation point TA. Using rare-earth 3d-transition metal ferrimagnetic compounds where net magnetic moment is nonzero at TA, the field-driven DW mobility remarkably enhances up to 20 km/sT. The collective coordinate approach generalized for ferrimagnets and atomistic spin model simulations show that this remarkable enhancement is a consequence of antiferromagnetic spin dynamics at TA. Our finding allows us to investigate the physics of antiferromagnetic spin dynamics and highlights the importance of tuning of the angular momentum compensation point of ferrimagnets, which could be a key towards ferrimagnetic spintronics.

cond-mat.mtrl-sci↗

arXiv2017

Spin-orbit effective fields in Pt/GdFeCo bilayers

In the increasing interests on spin-orbit torque (SOT) with various magnetic materials, we investigated SOT in rare earth-transition metal ferrimagnetic alloys. The harmonic Hall measurements were performed in Pt/GdFeCo bilayers to quantify the effective fields resulting from the SOT. It is found that the damping-like torque rapidly increases near the magnetization compensation temperature TM of the GdFeCo, which is attributed to the reduction of the net magnetic moment.

cond-mat.mtrl-sci↗