Despite remarkable advances in medical large language models(LLMs),their deployment in real clinical settings remains impractical due to prohibitive computational requirements and privacy regulations that restrict clo...Despite remarkable advances in medical large language models(LLMs),their deployment in real clinical settings remains impractical due to prohibitive computational requirements and privacy regulations that restrict cloud-based solutions.Small LLMs(sLLMs)offer a promising alternative for on-premise deployment,yet they require domain-specific fine-tuning that still exceeds the hardware capacity of most healthcare institutions.Furthermore,the impact of multilingual data composition on medical sLLM performance remains poorly understood.We present a resource-efficient fine-tuning pipeline that integrates Quantized Low-Rank Adaptation(QLoRA),Fully Sharded Data Parallelism(FSDP),and Sequence Packing,validated across two model scales:MedGemma 4B for efficiency analysis and LLaMA 3.370B for data balance experiments.Our approach achieves 58.3%reduction in video random access memory(VRAM)usage(from 48 GB to 20 GB)and 5×training speedup on MedGemma 4B using NVIDIA L40s GPUs.Critically,experiments on LLaMA 3.370B reveal that English-heavy data mixing(10:3 ratio)degrades Korean medical law performance by 1.23 percentage points while providing only marginal English gains(+1.49 pp),demonstrating catastrophic forgetting in multilingual medical fine-tuning.Our work provides three contributions:(1)a practical fine-tuning pipeline operable within 20 GB VRAM,(2)empirical evidence that data balance—not volume—determines multilingual medical QA performance,and(3)actionable guidelines for deploying medical sLLMs in non-English clinical environments.展开更多
Anomalous structural characteristics of the so-called first sharp diffraction peak (FSDP) that arises in the total static structure functions of network-forming glasses and liquids at around 1-2 A-1 have been rev...Anomalous structural characteristics of the so-called first sharp diffraction peak (FSDP) that arises in the total static structure functions of network-forming glasses and liquids at around 1-2 A-1 have been reviewed and discussed in details. Unlike other peaks in the static structure functions, the FSDP has anomalous dependencies on temperature, pressure and composition. Despite the fact that the FSDP is considered as a signature of intermediate range order (IRO) in network-forming glasses and liquids, its structural origin remains unclear and till now, it forms a subject of debate. A brief account for some anomalous characteristics of the FSDP followed by the different controversial interpretations about its structural origin has been reviewed and discussed. Some of the interpretations that seem to be inconsistent with recent experimental results have been ruled out. The most likely structural origins for the occurrence of the FSDP have been highlighted and discussed in details.展开更多
基金supported by a grant of the project for‘Research and Development for Enhancing Infectious Disease Response Capacity in Medical&Healthcare settings’,funded by the Korea Disease Control and Prevention Agency,the Ministry of Health&Welfare,Republic of Korea(grant number:RS-2025-02310471)supported by‘Research Base Construction Fund Support Program’funded by Jeonbuk National University in 2025.
摘要Despite remarkable advances in medical large language models(LLMs),their deployment in real clinical settings remains impractical due to prohibitive computational requirements and privacy regulations that restrict cloud-based solutions.Small LLMs(sLLMs)offer a promising alternative for on-premise deployment,yet they require domain-specific fine-tuning that still exceeds the hardware capacity of most healthcare institutions.Furthermore,the impact of multilingual data composition on medical sLLM performance remains poorly understood.We present a resource-efficient fine-tuning pipeline that integrates Quantized Low-Rank Adaptation(QLoRA),Fully Sharded Data Parallelism(FSDP),and Sequence Packing,validated across two model scales:MedGemma 4B for efficiency analysis and LLaMA 3.370B for data balance experiments.Our approach achieves 58.3%reduction in video random access memory(VRAM)usage(from 48 GB to 20 GB)and 5×training speedup on MedGemma 4B using NVIDIA L40s GPUs.Critically,experiments on LLaMA 3.370B reveal that English-heavy data mixing(10:3 ratio)degrades Korean medical law performance by 1.23 percentage points while providing only marginal English gains(+1.49 pp),demonstrating catastrophic forgetting in multilingual medical fine-tuning.Our work provides three contributions:(1)a practical fine-tuning pipeline operable within 20 GB VRAM,(2)empirical evidence that data balance—not volume—determines multilingual medical QA performance,and(3)actionable guidelines for deploying medical sLLMs in non-English clinical environments.
摘要Anomalous structural characteristics of the so-called first sharp diffraction peak (FSDP) that arises in the total static structure functions of network-forming glasses and liquids at around 1-2 A-1 have been reviewed and discussed in details. Unlike other peaks in the static structure functions, the FSDP has anomalous dependencies on temperature, pressure and composition. Despite the fact that the FSDP is considered as a signature of intermediate range order (IRO) in network-forming glasses and liquids, its structural origin remains unclear and till now, it forms a subject of debate. A brief account for some anomalous characteristics of the FSDP followed by the different controversial interpretations about its structural origin has been reviewed and discussed. Some of the interpretations that seem to be inconsistent with recent experimental results have been ruled out. The most likely structural origins for the occurrence of the FSDP have been highlighted and discussed in details.