Searcharxiv⌕ Search

arXiv subjects

Andrea De Angelis

Publications and source records attributed to Andrea De Angelis.

2 recordsLinked to original sources

Lara Mini Base: Language Diversity as a Resource, Not a Trade-off

General-purpose language models are trained mostly on English: over open models we review, at most, non-English pretraining data makes up only a 12% share. Models built for European languages give them far more, but score well below general-purpose models of the same size. A common concern is that, in small models, languages compete for capacity. We present Lara Mini base, a model trained from scratch on 6T tokens in English and 44 other languages, which together cover all 24 official languages of the European Union. As training progresses, the 44 other languages together come to outweigh English. To give these languages enough capacity at low cost, Lara Mini base uses a mixture of experts architecture with 7.2B total and 0.6B active parameters. Among open base models with up to 2B active parameters, it achieves the highest average score across standard English, multilingual, math, and code benchmarks, ahead of models trained on 4 to 6 times as many tokens. Its margin is largest on multilingual benchmarks in the languages it supports and on math benchmarks. It supports contexts of up to 128k tokens. We find that our MoE architecture reduces capacity competition between languages. For a single user on one GPU, without quantization, it generates text at least 1.8 times faster than any model with a higher average score. We release the weights of Lara Mini base under the CC BY-NC-SA 4.0 license. In addition, we will release T-HellaSwag, a higher-quality machine translation of the commonsense reasoning benchmark HellaSwag into 110 language variants.

cs.CL↗

Alaska: A Flexible Benchmark for Data Integration Tasks

Data integration is a long-standing interest of the data management community and has many disparate applications, including business, science and government. We have recently witnessed impressive results in specific data integration tasks, such as Entity Resolution, thanks to the increasing availability of benchmarks. A limitation of such benchmarks is that they typically come with their own task definition and it can be difficult to leverage them for complex integration pipelines. As a result, evaluating end-to-end pipelines for the entire data integration process is still an elusive goal. In this work, we present Alaska, the first benchmark based on real-world dataset to support seamlessly multiple tasks (and their variants) of the data integration pipeline. The dataset consists of ~70k heterogeneous product specifications from 71 e-commerce websites with thousands of different product attributes. Our benchmark comes with profiling meta-data, a set of pre-defined use cases with diverse characteristics, and an extensive manually curated ground truth. We demonstrate the flexibility of our benchmark by focusing on several variants of two crucial data integration tasks, Schema Matching and Entity Resolution. Our experiments show that our benchmark enables the evaluation of a variety of methods that previously were difficult to compare, and can foster the design of more holistic data integration solutions.

cs.DB↗