arXiv subjects

Nicolay J. Hammer

Publications and source records attributed to Nicolay J. Hammer.

6 recordsLinked to original sources

arXiv2025

Llama-GENBA-10B: A Trilingual Large Language Model for German, English and Bavarian

We present Llama-GENBA-10B, a trilingual foundation model addressing English-centric bias in large language models. Built on Llama 3.1-8B and scaled to 10B parameters, Llama-GENBA-10B is continuously pretrained on 164B tokens (82B English, 82B German, and 80M Bavarian), balancing resources while preventing English dominance. Targeted at the German NLP community, the model also promotes Bavarian as a low-resource language. Development tackled four challenges: (1) curating a multilingual corpus despite Bavarian scarcity, (2) creating a unified tokenizer for English, German, and Bavarian, (3) optimizing architecture and language-ratio hyperparameters for cross-lingual transfer, and (4) establishing the first standardized trilingual evaluation suite by translating German benchmarks into Bavarian. Evaluations show that Llama-GENBA-10B achieves strong cross-lingual performance, with the fine-tuned variant surpassing Apertus-8B-2509 and gemma-2-9b in Bavarian and establishing itself as the best model in its class for this language, while also outperforming EuroLLM in English and matching its results in German. Training on the Cerebras CS-2 demonstrated efficient large-scale multilingual pretraining with documented energy use, offering a blueprint for inclusive foundation models that integrate low-resource languages.

cs.CL↗

arXiv2018

Exploiting the Space Filling Curve Ordering of Particles in the Neighbour Search of Gadget3

Gadget3 is nowadays one of the most frequently used high performing parallel codes for cosmological hydrodynamical simulations. Recent analyses have shown t\ hat the Neighbour Search process of Gadget3 is one of the most time-consuming parts. Thus, a considerable speedup can be expected from improvements of the u\ nderlying algorithms. In this work we propose a novel approach for speeding up the Neighbour Search which takes advantage of the space-filling-curve particle ordering. Instead of performing Neighbour Search for all particles individually, nearby active particles can be grouped and one single Neighbour Search can be performed to obta\ in a common superset of neighbours. Thus, with this approach we reduce the number of searches. On the other hand, tree walks are performed within a larger searching radius. There is an optimal size of grouping that maximize the speedup, which we found by numerical experiments. We tested the algorithm within the boxes of the Magneticum project. As a result we obtained a speedup of $1.65$ in the Density and of $1.30$ in the Hydrodynamics computation, respectively, and a total speedup of $1.34.$

astro-ph.IM↗

arXiv2018

A web portal for hydrodynamical, cosmological simulations

This article describes a data center hosting a web portal for accessing and sharing the output of large, cosmological, hydro-dynamical simulations with a broad scientific community. It also allows users to receive related scientific data products by directly processing the raw simulation data on a remote computing cluster. The data center has a multi-layer structure: a web portal, a job control layer, a computing cluster and a HPC storage system. The outer layer enables users to choose an object from the simulations. Objects can be selected by visually inspecting 2D maps of the simulation data, by performing highly compounded and elaborated queries or graphically by plotting arbitrary combinations of properties. The user can run analysis tools on a chosen object. These services allow users to run analysis tools on the raw simulation data. The job control layer is responsible for handling and performing the analysis jobs, which are executed on a computing cluster. The innermost layer is formed by a HPC storage system which hosts the large, raw simulation data. The following services are available for the users: (I) {\sc ClusterInspect} visualizes properties of member galaxies of a selected galaxy cluster; (II) {\sc SimCut} returns the raw data of a sub-volume around a selected object from a simulation, containing all the original, hydro-dynamical quantities; (III) {\sc Smac} creates idealised 2D maps of various, physical quantities and observables of a selected object; (IV) {\sc Phox} generates virtual X-ray observations with specifications of various current and upcoming instruments.

astro-ph.IM↗

arXiv2017

Performance Optimisation of Smoothed Particle Hydrodynamics Algorithms for Multi/Many-Core Architectures

We describe a strategy for code modernisation of Gadget, a widely used community code for computational astrophysics. The focus of this work is on node-level performance optimisation, targeting current multi/many-core IntelR architectures. We identify and isolate a sample code kernel, which is representative of a typical Smoothed Particle Hydrodynamics (SPH) algorithm. The code modifications include threading parallelism optimisation, change of the data layout into Structure of Arrays (SoA), auto-vectorisation and algorithmic improvements in the particle sorting. We obtain shorter execution time and improved threading scalability both on Intel XeonR ($2.6 \times$ on Ivy Bridge) and Xeon PhiTM ($13.7 \times$ on Knights Corner) systems. First few tests of the optimised code result in $19.1 \times$ faster execution on second generation Xeon Phi (Knights Landing), thus demonstrating the portability of the devised optimisation solutions to upcoming architectures.

cs.DC↗

arXiv2016

The world's largest turbulence simulations

Understanding turbulence is critical for a wide range of terrestrial and astrophysical applications. Here we present first results of the world's highest-resolution simulation of turbulence ever done. The current simulation has a grid resolution of 10048^3 points and was performed on 65536 compute cores on SuperMUC at the Leibniz Supercomputing Centre (LRZ). We present a scaling test of our modified version of the FLASH code, which updates the hydrodynamical equations in less than 3 micro seconds per cell per time step. A first look at the column density structure of the 10048^3 simulation is presented and a detailed analysis is provided in a forthcoming paper.

astro-ph.SR↗

arXiv2010

An axis-free overset grid in spherical polar coordinates for simulating 3D self-gravitating flows

A type of overlapping grid in spherical coordinates called the Yin-Yang grid is successfully implemented into a 3D version of the explicit Eulerian grid-based code PROMETHEUS including self-gravity. The modified code successfully passed several standard hydrodynamic tests producing results which are in very good agreement with analytic solutions. Moreover, the solutions obtained with the Yin-Yang grid exhibit no peculiar behaviour at the boundary between the two grid patches. The code has also been successfully used to model astrophysically relevant situations, namely equilibrium polytropes, a Taylor-Sedov explosion, and Rayleigh-Taylor instabilities. According to our results, the usage of the Yin-Yang grid greatly enhances the suitability and efficiency of 3D explicit Eulerian codes based on spherical polar coordinates for astrophysical flows.

astro-ph.IM↗