-
Introducing cosmosGPT: Monolingual Training for Turkish Language Models
Authors:
H. Toprak Kesgin,
M. Kaan Yuce,
Eren Dogan,
M. Egemen Uzun,
Atahan Uz,
H. Emre Seyrek,
Ahmed Zeer,
M. Fatih Amasyali
Abstract:
The number of open source language models that can produce Turkish is increasing day by day, as in other languages. In order to create the basic versions of such models, the training of multilingual models is usually continued with Turkish corpora. The alternative is to train the model with only Turkish corpora. In this study, we first introduce the cosmosGPT models that we created with this alter…
▽ More
The number of open source language models that can produce Turkish is increasing day by day, as in other languages. In order to create the basic versions of such models, the training of multilingual models is usually continued with Turkish corpora. The alternative is to train the model with only Turkish corpora. In this study, we first introduce the cosmosGPT models that we created with this alternative method. Then, we introduce new finetune datasets for basic language models to fulfill user requests and new evaluation datasets for measuring the capabilities of Turkish language models. Finally, a comprehensive comparison of the adapted Turkish language models on different capabilities is presented. The results show that the language models we built with the monolingual corpus have promising performance despite being about 10 times smaller than the others.
△ Less
Submitted 26 April, 2024;
originally announced April 2024.
-
Türkçe Dil Modellerinin Performans Karşılaştırması Performance Comparison of Turkish Language Models
Authors:
Eren Dogan,
M. Egemen Uzun,
Atahan Uz,
H. Emre Seyrek,
Ahmed Zeer,
Ezgi Sevi,
H. Toprak Kesgin,
M. Kaan Yuce,
M. Fatih Amasyali
Abstract:
The developments that language models have provided in fulfilling almost all kinds of tasks have attracted the attention of not only researchers but also the society and have enabled them to become products. There are commercially successful language models available. However, users may prefer open-source language models due to cost, data privacy, or regulations. Yet, despite the increasing number…
▽ More
The developments that language models have provided in fulfilling almost all kinds of tasks have attracted the attention of not only researchers but also the society and have enabled them to become products. There are commercially successful language models available. However, users may prefer open-source language models due to cost, data privacy, or regulations. Yet, despite the increasing number of these models, there is no comprehensive comparison of their performance for Turkish. This study aims to fill this gap in the literature. A comparison is made among seven selected language models based on their contextual learning and question-answering abilities. Turkish datasets for contextual learning and question-answering were prepared, and both automatic and human evaluations were conducted. The results show that for question-answering, continuing pretraining before fine-tuning with instructional datasets is more successful in adapting multilingual models to Turkish and that in-context learning performances do not much related to question-answering performances.
△ Less
Submitted 25 April, 2024;
originally announced April 2024.
-
Develo** and Evaluating Tiny to Medium-Sized Turkish BERT Models
Authors:
Himmet Toprak Kesgin,
Muzaffer Kaan Yuce,
Mehmet Fatih Amasyali
Abstract:
This study introduces and evaluates tiny, mini, small, and medium-sized uncased Turkish BERT models, aiming to bridge the research gap in less-resourced languages. We trained these models on a diverse dataset encompassing over 75GB of text from multiple sources and tested them on several tasks, including mask prediction, sentiment analysis, news classification, and, zero-shot classification. Despi…
▽ More
This study introduces and evaluates tiny, mini, small, and medium-sized uncased Turkish BERT models, aiming to bridge the research gap in less-resourced languages. We trained these models on a diverse dataset encompassing over 75GB of text from multiple sources and tested them on several tasks, including mask prediction, sentiment analysis, news classification, and, zero-shot classification. Despite their smaller size, our models exhibited robust performance, including zero-shot task, while ensuring computational efficiency and faster execution times. Our findings provide valuable insights into the development and application of smaller language models, especially in the context of the Turkish language.
△ Less
Submitted 26 July, 2023;
originally announced July 2023.
-
Galactic chemical evolution]{Galactic chemical evolution of the solar neighborhood, solar twins and exoplanet indicators
Authors:
Charles R. Cowley,
Kutluay Yüce
Abstract:
Galactic chemical evolution (GCE), solar analogues or twins, and peculiarities of the solar composition with respect to the twins are inextricably related. We examine GCE parameters from the literature and present newly derived values using a quadratic fit that gives zero for a Solar age (i.e., 4.6 Gyr). We show how the GCE parameters may be used not only to "correct" abundances to the solar age,…
▽ More
Galactic chemical evolution (GCE), solar analogues or twins, and peculiarities of the solar composition with respect to the twins are inextricably related. We examine GCE parameters from the literature and present newly derived values using a quadratic fit that gives zero for a Solar age (i.e., 4.6 Gyr). We show how the GCE parameters may be used not only to "correct" abundances to the solar age, but to predict relative elemental abundances as a function of age. We address the question of whether the solar abundances are depleted in refractories and enhanced in volatiles and find that the answer is sensitive to the selection of a representative standard. The best quality data sets do not support the notion that the Sun is depleted in refractories and enhanced in volatiles. A simple model allows us to estimate the amount of refractory-rich material missing from the Sun or alternately added to the average solar twin. The model gives between zero and 1.4 earth masses.
△ Less
Submitted 10 March, 2022;
originally announced March 2022.
-
Modeling stellar abundance patterns resulting from the addition of earthlike planetary material
Authors:
Charles R. Cowley,
Donald J. Bord,
Kutluay Yuce
Abstract:
The literature on precision differential abundances (PDAs) in stars is extensive. Surveys include sun-like stars in the solar neighborhood, binary systems, and Galactic clusters. Numerous references as well as a discussion of relevant mechanisms may be found in papers by Ramirez, et al. (2019) and Nagar, et al. (2020). A strong impetus for this work is the probability that the abundances have been…
▽ More
The literature on precision differential abundances (PDAs) in stars is extensive. Surveys include sun-like stars in the solar neighborhood, binary systems, and Galactic clusters. Numerous references as well as a discussion of relevant mechanisms may be found in papers by Ramirez, et al. (2019) and Nagar, et al. (2020). A strong impetus for this work is the probability that the abundances have been influenced by exoplanetary systems and their evolution.
We calculate the resulting differential abundances ([El/H]) assuming a given amount of material with the composition of the bulk earth (Wang, et al. 2018) was added to the stellar convection zone of a dwarf G-type star. The mass of the convection zone is uncertain and variable, depending on the spectral type. Here, we assume a mass of $5\cdot 10^{31}$ gm for the stellar convection zone (SCZ). This is 0.025 $M_\odot$ (Pinsonneault, et al. 2001, Chambers, 2010). For other SCZ masses, the parameters must be adjusted accordingly. In general, the sunlike star will not have exactly the solar composition. This contingency is roughly taken into account in our model. An appendix discusses issues of volatility and condensation temperature.
△ Less
Submitted 25 January, 2021;
originally announced January 2021.
-
HD 196390: A tight correlation of differential abundances with condensation temperature
Authors:
Charles R. Cowley,
Donald J. Bord,
Kutluay Yüce
Abstract:
Bedell et al. (2018) give precision differential abundances for 79 mostly G-dwarf stars. We correct these abundances for Galactic chemical evolution in a manner similar to that used by these authors but with parameters derived from linear fits to plots of [El/H] vs. age in lieu of [El/Fe]. We examine the resulting abundances for correlations with the 50% condensation temperature using values from…
▽ More
Bedell et al. (2018) give precision differential abundances for 79 mostly G-dwarf stars. We correct these abundances for Galactic chemical evolution in a manner similar to that used by these authors but with parameters derived from linear fits to plots of [El/H] vs. age in lieu of [El/Fe]. We examine the resulting abundances for correlations with the 50% condensation temperature using values from both Lodders (2003) and Wood et al. (2019), and compare with the results of Bedell et al. HD 196390 is distinct in having the most significant correlation of the 79-star sample. We report statistics for a subset of stars with lower significance, but of some interest.
△ Less
Submitted 6 April, 2020; v1 submitted 31 March, 2020;
originally announced March 2020.
-
Differential abundance analysis of Procyon and Theta Sculptoris: Comparison with abundance patterns of solar-like pairs
Authors:
C. R. Cowley,
K. Yüce,
D. J. Bord
Abstract:
The precision differential abundance (PDA) technique is applied to the mid-F stars Procyon and $θ$ Scl using spectra from the ESO UVESPOP library. We relate PDA patterns to endogenous processes related to condensation or to exogenous processes connected to Galactic chemical evolution (GCE). We employ one-dimensional LTE models, but emphasize the use of weaker lines ($\leq$ 20 mÅ) than are typicall…
▽ More
The precision differential abundance (PDA) technique is applied to the mid-F stars Procyon and $θ$ Scl using spectra from the ESO UVESPOP library. We relate PDA patterns to endogenous processes related to condensation or to exogenous processes connected to Galactic chemical evolution (GCE). We employ one-dimensional LTE models, but emphasize the use of weaker lines ($\leq$ 20 mÅ) than are typically used in such studies. We compare our results with PDAs of solar-type stars. Abundances and PDAs are determined for 28 elements: C, N, O, Na, Mg, Al, Si, S, Ca, Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Y, Zr, Ba, La, Ce, Nd, Sm, Eu, and Gd. A plot of PDAs ($θ$ Scl minus Procyon) vs. $Z$ shows a highly significant correlation. Moreover, local substructure of the plot for the elements Ca-Zn and neutron-addition elements is similar to that which can be found for solar twins. Our PDA vs. $Z$ plot structural similarity to plots that can be made from the extensive work of Bedell et al. (2018). That PDA structure and substructure is clearly a function of age.
△ Less
Submitted 13 December, 2019;
originally announced December 2019.
-
Carbon deficiencies in the primaries of some classical Algols
Authors:
C. Ibanoglu,
A. Dervisoglu,
O. Cakirli,
E. Sipahi,
K. Yuce
Abstract:
The equivalent widths of C II $λ$ 4267 Åline were measured for the mass-gaining primary stars of the 18 Algol-type binary systems. The comparison of the EWs of the gainers with those of the single standard stars having the same effective temperature and luminosity class clearly indicates that they are systematically smaller than those of the standard stars. The primary components of the classical…
▽ More
The equivalent widths of C II $λ$ 4267 Åline were measured for the mass-gaining primary stars of the 18 Algol-type binary systems. The comparison of the EWs of the gainers with those of the single standard stars having the same effective temperature and luminosity class clearly indicates that they are systematically smaller than those of the standard stars. The primary components of the classical Algols, located in the main-sequence band of the HR diagram, appear to be C poor stars. We estimate $ [N_{C} /N_{tot}] $ relative to the Sun as -1.91 for GT Cep, -1.88 for AU Mon and -1.41 for TU Mon, indicating poorer C abundance. An average differential carbon abundance has been estimated to be -0.82 dex relative to the Sun and -0.54 dex relative to the main-sequence standard stars. This result is taken to be an indication of the transferring material from the evolved less-massive secondary components to the gainers such that the CNO cycle processed material changed the original abundance of the gainers. There appear to be relationships between the EWs of C II $λ$ 4267 Åline and the rates orbital period increase and mass transfer in some Algols. As the mass transfer rate increases the EW of the C II line decreases, which indicates that accreted material has not been completely mixed yet in the surface layers of the gainers. This result supports the idea of mixing as an efficient process to remove the abundance anomaly built up by accretion. Chemical evolution of the classical Algol-type systems may lead to constrains on the initial masses of the less massive, evolved, mass-losing stars.
△ Less
Submitted 9 September, 2011;
originally announced September 2011.
-
Wavelengths and oscillator strengths of Xe II from the UVES spectra of four HgMn stars
Authors:
K. Yüce,
F. Castelli,
S. Hubrig
Abstract:
In spite of large overabundances of Xe II observed in numerous mercury-manganese (HgMn) stars, Xe II oscillator strengths are only available for a very limited number of transitions. As a consequence, several unidentified lines in the spectra of HgMn stars could be due to Xe II. In addition, some predicted Xe II lines are redshifted by about 0.1 A from stellar unidentified lines, raising the quest…
▽ More
In spite of large overabundances of Xe II observed in numerous mercury-manganese (HgMn) stars, Xe II oscillator strengths are only available for a very limited number of transitions. As a consequence, several unidentified lines in the spectra of HgMn stars could be due to Xe II. In addition, some predicted Xe II lines are redshifted by about 0.1 A from stellar unidentified lines, raising the question about the wavelength accuracy of the Xe II line data available in the literature. For these reasons we investigated the Xe II lines lying in the 3900-4521 A, 4769-7542 A, and 7660-8000 A spectral ranges of four well-studied HgMn stars. We compared the Xe II wavelengths listed in the NIST database with the position of the lines observed in the high-resolution UVES spectrum of the xenon-overabundant, slowly rotating HgMn stars HR 6000, and we modified them when needed. We derived astrophysical oscillator strengths for all the Xe II observed lines and compared them with the literature values, when available. In this framework, we performed a complete abundance analysis of HD 71066, while we relied on our previous works for the other stars. We find that all the lines with wavelengths related to the 6d and 7s energy levels have a corresponding unidentified spectral line, blueshifted by the same quantity of about 0.1 A in all the four stars, so that we identified these lines as coming from Xe II and modified their NIST wavelength value according to the observed stellar value. We find that the Xe II stellar oscillator strengths may differ from one star to another from 0.0 dex to 0.3 dex. We adopted the average of the oscillator strengths derived from the four stars as final astrophysical oscillator strength.
△ Less
Submitted 19 January, 2011;
originally announced January 2011.
