To the Editor, We read with interest Aali et al.’s study introducing MIMIC-IV-BHC and a benchmark of large language models (LLMs) for brief hospital-course (BHC) summaries.1 The work makes a timely contribution by releasing resources and reporting baselines across context lengths and adaptation strategies. We suggest several additions that could further strengthen the benchmark’s clinical validity, reproducibility, and generalizability. In sum, Aali et al. provide an important foundation for BHC summarization benchmarking. Augmenting evaluation with factuality/safety metrics, more informative reader-study reporting, robustness stress tests, and external/multilingual validation would further align the benchmark with clinical veracity and translational readiness. Jose Zablah (Methodology, Writing–original draft), Yolly Molina (Validation, Writing–original draft), and Antonio Garcia Loureiro (Conceptualization, Writing–review & editing) None declared. None declared. No new datasets or code was generated for this article. Data discussed here are publicly hosted as follows: the MIMIC-IV-Ext-BHC labeled notes resource for hospital-course summarization (Version 1.2.0; https://doi.org/10.13026/5gte-bv70) is available to credentialed users on PhysioNet; it derives from MIMIC-IV-Note (Version 2.2; https://doi.org/10.13026/1n74-ne17) and is linkable to the MIMIC-IV clinical database (Version 3.1; https://doi.org/10.13026/kpb9-mt58). Access policies (DUA and CITI training) are detailed on the PhysioNet project pages. The JAMIA article introducing the benchmark is available at https://doi.org/10.1093/jamia/ocae312, and the accompanying benchmarking scripts are maintained at https://github.com/StanfordMIMI/clin-bhc-summ. Evaluation tools referenced in the text are openly available: QAFactEval (https://github.com/salesforce/QAFactEval) and finally CheckList https://github.com/marcotcr/checklist).

Summary Emerging new architectures used in High Performance Computing require new research to adapt and optimise algorithms to them. As part of this effort, we propose the new AXC format to improve the performance of the SpMV product for the Intel Xeon Phi coprocessor. The performance of the OpenCL kernel, based on our new format, is compared with three very different and high efficient sparse matrix formats, ie, CSR, ELLR‐T, and K1. We perform tests with most of the matrices from the Williams collection used to test SpMV kernels for GPUs architectures in several related works. The numerical results show that the AXC format is more robust to spatial indirections proper of sparse matrices and prefers matrices with low variability amongst their rows' population, very much like matrices originated by FEM codes. The Conjugate Gradient (CG) is implemented in OpenCL using all the formats in this work to expose strengths and weaknesses of the formats in a real application. The CG implementation shows that the AXC has the fastest conversion time and its coherent with the numerical results generated by the SpMV tests, and that the format has a slower memory operations time due to an extra step required by the format and its larger memory footprint.

Nowadays, there are several open-source solutions for building private, public and even hybrid clouds such as Eucalyptus, Apache Cloud Stack and Open Stack. KVM is one of the supported hypervisors for these cloud platforms. Different KVM configurations are being supplied by these platforms and, in some cases, a subset of CPU features are being presented to guest systems, providing a basic abstraction of the underlying CPU. One of the reasons for limiting the features of the Virtual CPU is to guarantee the guest compatibility with different hardware in heterogeneous environments. However, in a large number of situations, the cloud is deployed on an homogeneous set of hosts. In these cases, this limitation can affect the performance of applications being executed in guest systems. In this paper, we have analyzed the architecture, the KVM setup, and the performance of the Virtual Machines deployed by three popular cloud management platforms: Eucalyptus, Apache Cloud Stack and Open Stack, employing a representative set of applications.

Currently, there are an increasing number of open-source solutions for building Clouds. The performance of the Virtual Machines running in these Clouds is a key point. The way the Virtual Machines are created in Clouds can have important effects upon their disk I/O operations. We have selected CloudStack platform to study the disk I/O performance for KVM Virtual Machines.

An efficient implementation of the density-gradient (DG) approach for the finite element and finite difference methods and its application in drift-diffusion (D-D) simulations is described in detail. The new, second-order differential (SOD) scheme is compatible with relatively coarse grids even for large density variations thus applicable to device simulations with complex 3-D geometries. Test simulations of a 1-D metal-oxide semiconductor diode demonstrate that the DG approach discretized using our SOD scheme can be accurately calibrated against Schrödinger-Poisson calculations exhibiting lower discretization error than the previous schemes when using coarse grids and the same results for very fine meshes. 3-D test D-D simulations using the finite element method are performed on two devices: a 10 nm gate length double gate metal-oxide-semiconductor field-effect transistor (MOSFET) and a 40 nm gate length Tri-Gate fin field-effect transistor (FinFET). In 3-D D-D simulations, the SOD scheme is able to converge to physical solutions at high voltages even if the previous schemes fail when using the same mesh and equivalent conditions. The quantum corrected D-D simulations using the SOD scheme also converge with an atomistic mesh used for the 10 nm double gate MOSFET saving computational resources and can be accurately calibrated against the results from non-equilibrium Green's functions approach. Finally, the simulated ID-VGcharacteristics for the 40 nm gate length Tri-Gate are in an excellent agreement with experimental data.

The next substitution of the Enabling Grids for E-sciencE project (EGEE) in 2010 by the European Grid Initiative (EGI), where grid infrastructure of each country will be run by National Grid Initiatives (NGI), is giving a boost to the NGI development. In this context, the Spanish National Grid Initiative (es-NGI) is being developed by the e-science Spanish network. The es-NGI is developing virtual organizations where common area applications are associated. In this context, the MOSFET virtual organisation (VO-MOSFET) was born in 2009 to perform semiconductor device simulations using the es-NGI infrastructure. This virtual organization is supported by the es-NGI resource centres and it is developing a job submission and monitoring system independent of the grid middleware. In this paper a general description of the VO-MOSFET and some application level utilities of the job submission system are presented. Furthermore, a gridification example of a nanoelectronic simulation is presented proving the grid benefits for the field of nanoelectronic simulation.

This paper presents the optimization and evaluation of parallel I/O for the BIPS3D parallel irregular application, a 3-dimensional simulation of BJT and HBT bipolar devices. The parallel version of BIPS3D employs Metis, a library for partitioning graphs, finite element meshes, or sparse matrices. First, we show how the partitioning information provided by Metis can be used in order to improve the performance of parallel I/O. Second, we propose a novel technique, called Interval Data Grouping (IDG), which exploits the data replication of mesh nodes for optimizing the scheduling of the parallel file operations. Finally, we evaluate the parallel I/O version of BIPS3D for various existing parallel I/O techniques and present an in-depth analysis of the IDG performance.