291 lines
17 KiB
Markdown
291 lines
17 KiB
Markdown
---
|
||
license: apache-2.0
|
||
language:
|
||
- ca
|
||
- es
|
||
- en
|
||
base_model: BSC-LT/salamandra-7b
|
||
tags:
|
||
- valencian
|
||
- catalan
|
||
- spanish
|
||
- english
|
||
- text-generation
|
||
- alia
|
||
- gplsi
|
||
datasets:
|
||
- gplsi/alia_dogv
|
||
- gplsi/alia_les_corts
|
||
- gplsi/alia_amic
|
||
- gplsi/alia_boua
|
||
- gplsi/alia_tourism
|
||
library_name: transformers
|
||
pipeline_tag: text-generation
|
||
---
|
||
|
||
# Aitana-7B-S-base
|
||
|
||
**Aitana-7B-S-base** is a generative language model from the **Aitana family**, developed by the [GPLSI (Language and Information System Group)](https://gplsi.dlsi.ua.es/) at the University of Alicante and [Language Modeling Group at Barcelona Supercomputing Center](https://www.bsc.es/research-development/research-areas/cognitive-computing/language-modeling). This model is based on [BSC-LT/salamandra-7b](https://huggingface.co/BSC-LT/salamandra-7b) and has been continuously pre-trained on multilingual data (Valencian, Spanish, and English) to improve representation of Valencian and Catalan languages.
|
||
|
||
## Table of Contents
|
||
|
||
- [Model Description](#model-description)
|
||
- [Evaluation](#evaluation)
|
||
- [Training Data](#training-data)
|
||
- [Intended Uses](#intended-uses)
|
||
- [How to Use](#how-to-use)
|
||
- [Additional Information](#additional-information)
|
||
|
||
## Model Description
|
||
|
||
| Property | Value |
|
||
|----------|-------|
|
||
| **Base Model** | [BSC-LT/salamandra-7b](https://huggingface.co/BSC-LT/salamandra-7b) |
|
||
| **Architecture** | Transformer decoder-only |
|
||
| **Parameters** | ~7.77B |
|
||
| **Languages** | Valencian, Spanish, English |
|
||
| **License** | Apache 2.0 |
|
||
|
||
Aitana-7B-S-base extends the multilingual Salamandra foundation with additional training on domain-specific Valencian, Spanish, and English data. The training emphasizes administrative, legal, and tourism domains.
|
||
|
||
## Training Data
|
||
|
||
This model was trained on the following ALIA datasets:
|
||
|
||
| Dataset ID | Name | Language | Source |
|
||
|------------|------|----------|--------|
|
||
| dc8 | dogv_va_2025 | Valencian | [gplsi/alia_dogv](https://huggingface.co/datasets/gplsi/alia_dogv) |
|
||
| dc9 | dogv_es_2025 | Spanish | [gplsi/alia_dogv](https://huggingface.co/datasets/gplsi/alia_dogv) |
|
||
| dc10 | corts_es_va_2025 | Spanish/Valencian | [gplsi/alia_les_corts](https://huggingface.co/datasets/gplsi/alia_les_corts) |
|
||
| dc11 | amic_va_2025 | Valencian | [gplsi/alia_amic](https://huggingface.co/datasets/gplsi/alia_amic) |
|
||
| dc12 | boua_va_2025 | Valencian | [gplsi/alia_boua](https://huggingface.co/datasets/gplsi/alia_boua) |
|
||
| dc13 | boua_es_2025 | Spanish | [gplsi/alia_boua](https://huggingface.co/datasets/gplsi/alia_boua) |
|
||
| dc14 | tourism_va_2025 | Valencian | [gplsi/alia_tourism](https://huggingface.co/datasets/gplsi/alia_tourism) |
|
||
| dc15 | tourism_es_2025 | Spanish | [gplsi/alia_tourism](https://huggingface.co/datasets/gplsi/alia_tourism) |
|
||
| dc16 | tourism_en_2025 | English | [gplsi/alia_tourism](https://huggingface.co/datasets/gplsi/alia_tourism) |
|
||
|-|alia_multilingual_parallel_sentences|Spanish/Valencian/English|[gplsi/alia_multilingual_parallel_sentences](https://huggingface.co/datasets/gplsi/alia_multilingual_parallel_sentences)|
|
||
|
||
### Data Sources
|
||
|
||
- **DOGV (Diari Oficial de la Generalitat Valenciana)**: Official communications of the Valencian Community including laws and public sector communications
|
||
- **Les Corts Valencianes**: Transcripts from the Valencian Parliament plenary sessions and committee meetings
|
||
- **AMIC**: Valencian language corpus
|
||
- **BOUA (Butlletí Oficial de la Universitat d'Alacant)**: Official University of Alicante documents including grants, regulations, and resolutions
|
||
- **Tourism**: Multilingual tourism domain content
|
||
|
||
## Intended Uses
|
||
|
||
This model can be used for:
|
||
|
||
- **Text generation** in Valencian, Spanish, and English
|
||
- **Fine-tuning** for specific downstream tasks
|
||
- **Domain adaptation** for administrative, legal, or tourism applications
|
||
|
||
> **Note**: Due to the formal register of training data (administrative and legal domains), generated text tends toward formal language.
|
||
|
||
## How to Use
|
||
|
||
### Transformers
|
||
|
||
```python
|
||
import torch
|
||
from transformers import pipeline, AutoTokenizer
|
||
|
||
model_id = "gplsi/Aitana-7B-S-base"
|
||
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
||
|
||
generator = pipeline(
|
||
"text-generation",
|
||
model=model_id,
|
||
tokenizer=tokenizer,
|
||
torch_dtype=torch.bfloat16,
|
||
device_map="auto",
|
||
)
|
||
|
||
# Valencian example
|
||
text = "Les corts valencianes han pres la decisió de"
|
||
result = generator(text, do_sample=True, top_k=10, max_new_tokens=100)
|
||
print(result[0]['generated_text'])
|
||
|
||
# Spanish example
|
||
text = "El turismo en la Comunidad Valenciana"
|
||
result = generator(text, do_sample=True, top_k=10, max_new_tokens=100)
|
||
print(result[0]['generated_text'])
|
||
```
|
||
|
||
|
||
## Evaluation
|
||
|
||
In the following table, we can see the results obtained with different benchmarks from [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) in comparison with the model used for continuous pre-training.
|
||
The results have been obtained from the model pre-trained; no instruction tuning or fine-tuning of any kind has been performed.
|
||
|
||
### Normalized score per language
|
||
|
||
| Language | Salamandra-7B | Aitana-7B-S-base |
|
||
|----------|----------|----------|
|
||
| **Spanish** | 0.248 | **0.26** |
|
||
| **Catalan** | 0.364 | **0.373** |
|
||
| **English** | 0.319 | **0.349** |
|
||
| **Valencian** | 0.663 | **0.664** |
|
||
|
||
### Valencian
|
||
|
||
|
||
#### Classification Benchmarks
|
||
|
||
| Dataset | Lang. | Task | Metric | Salamandra-7B | Aitana-7B-S-base |
|
||
|------------------------------|--------|----------------------------|-------------|---------------|-----------------------|
|
||
| XNLI | va |Natural Language Inference | acc | **0.496** | 0.495 |
|
||
|
||
|
||
#### Generation Benchmarks
|
||
|
||
| Dataset | Lang. | Task | Metric | Salamandra-7B | Aitana-7B-S-base |
|
||
|------------------------------|--------|----------------------------|-------------|---------------|-----------------------|
|
||
| Cocoteros | va |Reading Comprehension | bleu | 12.30 | **16.09** |
|
||
| Phrases ca-va | va-ca |Translation - Adaptation | bleu | **86.83** | 86.53 |
|
||
| Phrases va-ca | va-ca |Translation - Adaptation | bleu | **94.68** | 82.99 |
|
||
| Phrases va-es | va-es |Translation | bleu | 79.83 | **80.76** |
|
||
| Phrases es-va | es-va |Translation | bleu | 66.31 | **71.01** |
|
||
| Truthfulqa_va | va | Truthfulness | bleu_acc| 0.353 | **0.388** |
|
||
|
||
|
||
### Catalan
|
||
|
||
|
||
#### Classification Benchmarks
|
||
|
||
| Dataset | Lang. | Task | Metric | Salamandra-7B | Aitana-7B-S-base |
|
||
|------------------------------|--------|----------------------------|-------------|---------------|-----------------------|
|
||
| Belebele Cat_latn | ca | Reading Comprehension | acc | 0.51 | **0.546** |
|
||
| COPA | ca | Commonsense Reasoning | acc | 0.798 | **0.812** |
|
||
| XStoryCloze | ca | Commonsense Reasoning | acc | 0.75 | **0.767** |
|
||
| OpenBookQA | ca | Question Answering | acc | 0.366 | **0.376** |
|
||
| PAWS | ca | Paraphrasing | acc | **0.626** | 0.613 |
|
||
| PiQA | ca | Question Answering | acc | 0.702 | **0.725** |
|
||
| SiQA | ca | Question Answering | acc | 0.489 | **0.506** |
|
||
| ARC Easy | ca | Question Answering | acc | 0.726 | **0.73** |
|
||
| ARC Challenge | ca | Question Answering | acc | **0.47** | 0.459 |
|
||
| XNLI | ca | Natural Language Inference | acc | **0.504** | 0.494 |
|
||
| Teca | ca | Natural Language Inference | acc | **0.527** | 0.514. |
|
||
| WNLI | ca | Natural Language Inference | acc | 0.577 | **0.633** |
|
||
| Catcola | ca | Linguistic Acceptability | acc | **0.732** | 0.71
|
||
| Catalanqa | ca | Question Answering | F1 | **0.832** | 0.829 |
|
||
| Catalanqa | ca | Question Answering | exact match | 0.62 | **0.65** |
|
||
| Mgsm direct | ca | Math | exact match | 0.068 | **0.096** |
|
||
| Xquad | ca | Question Answering | exact match | **0.498** | 0.497 |
|
||
| Xquad | ca | Question Answering | F1 | 0.717 | **0.724** |
|
||
|
||
|
||
|
||
#### Generation Benchmarks
|
||
|
||
| Dataset | Lang. | Task | Metric | Salamandra-7B | Aitana-7B-S-base |
|
||
|------------------------------|--------|----------------------------|--------|----------------|-----------------------|
|
||
| Cabreu abstractive | ca | Summarization | bleu | 8.46 | **11.34** |
|
||
| Cabreu extractive | ca | Summarization | bleu | **44.62** | 41.73 |
|
||
| Cabreu extreme | ca | Summarization | bleu | 11.02 | **12.44** |
|
||
|
||
|
||
|
||
### Spanish
|
||
|
||
#### Classification Benchmarks
|
||
|
||
| Dataset | Lang. | Task | Metric | Salamandra-7B | Aitana-7B-S-base |
|
||
|------------------------------|--------|---------------------------|-------------|---------------|-----------------------|
|
||
| Belebele | es | Reading Comprehension | acc | 0.49 | **0.55** |
|
||
| PAWS | es | Paraphrasing | acc | **0.616** | 0.591 |
|
||
| XNLI | es | Natural Language Inference| acc | **0.462** | 0.447 |
|
||
| WNLI | es | Natural Language Inference| acc | **0.45** | **0.45** |
|
||
| XStoryCloze | es | Commonsense Reasoning | acc | 0.746 | **0.754** |
|
||
| Escola | es | Linguistic Acceptability | acc | - | - |
|
||
| Escola | es | Linguistic Acceptability | mcc | - | - |
|
||
| OpenbookQA | es | Question Answering | acc | - | - |
|
||
| MGSM Direct | es | Math | exact match | 0.064 | **0.084** |
|
||
| XQUAD | es | Question Answering | exact match | **0.51** | 0.509 |
|
||
| XQUAD | es | Question Answering | F1 | 0.746 | **0.754** |
|
||
|
||
|
||
|
||
#### Generation Benchmarks
|
||
|
||
| Dataset | Lang. | Task | Metric | Salamandra-7B | Aitana-7B-S-base |
|
||
|------------------------------|--------|---------------------|---------|----------------|-----------------------|
|
||
| Cocoteros | es |Reading Comprehension| bleu | 14.57 | **17.35** |
|
||
| XLSum | es | Summarization | bleu | 3.52 | **5.79** |
|
||
|
||
### English
|
||
|
||
#### Classification Benchmarks
|
||
|
||
| Dataset | Lang. | Task | Metric | Salamandra-7B | Aitana-7B-S-base |
|
||
|------------------------------|--------|----------------------------|-------------|---------------|-----------------------|
|
||
| Arc Challenge | en | Question Answering | acc | **0.53** | 0.529 |
|
||
| Arc Easy | en | Question Answering | acc | **0.822** | 0.816 |
|
||
| Belebele | en | Reading Comprehension | acc | **0.562** | 0.537 |
|
||
| PAWS | en | Paraphrasing | acc | **0.632** | 0.604 |
|
||
| XNLI | en | Natural Language Inference | acc | **0.474** | 0.472 |
|
||
| XStoryCloze | en | Commonsense Reasoning | acc | **0.796** | 0.79 |
|
||
| OpenBookQA | en | Question Answering | acc | 0.352 | **0.356** |
|
||
| PiQA | en | Question Answering | acc | 0.793 | **0.796** |
|
||
| Social iqa | en | Question Answering | acc | **0.509** | 0.508 |
|
||
| WNLI | en | Natural Language Inference | acc | 0.464 | **0.549** |
|
||
| MGSM Direct | en | Math | exact match | 0.264 | **0.564** |
|
||
| TriviaQA | en | Question Answering | exact match | 0.597 | **0.601** |
|
||
| CoLA | en | Linguistic Acceptability | mcc | **0.381** | 0.339 |
|
||
|
||
## Additional Information
|
||
|
||
### Author
|
||
|
||
The model has been developed by the **Language and [Information Systems Group (GPLSI)](https://gplsi.dlsi.ua.es/)**, the **[Centro de Inteligencia Digital (CENID)](https://cenid.es)**, and the [Language Modeling Group at Barcelona Supercomputing Center] (https://www.bsc.es/research-development/research-areas/cognitive-computing/language-modeling), all contributing to cutting-edge research in Natural Language Processing (NLP). GPLSI and CENID are part of the **[University of Alicante (UA)](https://www.ua.es/es/)**, while the Language Modeling Group operates within the [Barcelona Supercomputing Center](https://www.bsc.es/es).
|
||
|
||
|
||
### Funding
|
||
|
||
This work is funded by the **Ministerio para la Transformación Digital y de la Función Pública**, co-financed by the **EU – NextGenerationEU**, within the framework of the project *Desarrollo de Modelos ALIA*. This work has also been partially supported by Project HEART-NLP (PID2024-156263OB-C22).
|
||
|
||
|
||
### Acknowledgments
|
||
|
||
We would like to express our gratitude to all individuals and institutions that have contributed to the development of this work.
|
||
|
||
Special thanks to:
|
||
- [Language Modeling at Barcelona Supercomputing Center](https://www.bsc.es/research-development/research-areas/cognitive-computing/language-modeling)
|
||
- [Centro Vasco de Tecnología de la Lengua (HiTZ)](https://www.hitz.eus/es)
|
||
- [Centro Singular de Investigación en Tecnologías Inteligentes (CiTIUS)](https://citius.gal/)
|
||
- [Sistemas Inteligentes de Acceso a la Información (SINAI)](https://www.ujaen.es/investigacion-y-transferencia/grupos-de-investigacion/sistemas-inteligentes-de-acceso-la-informacion-sinai)
|
||
- [Instituto Universitario de Investigación Informática (IUII)](https://web.ua.es/es/iuii/)
|
||
- [Leonardo HPC System](https://leonardo-supercomputer.cineca.eu/)
|
||
- [European supercomputing ecosystem (EUROHPC)](https://www.eurohpc-ju.europa.eu/)
|
||
|
||
|
||
We also acknowledge the financial, technical, and scientific support of the **Ministerio para la Transformación Digital y de la Función Pública - Funded by EU – NextGenerationEU within the framework of the project Desarrollo de Modelos ALIA**, whose contribution has been essential to the completion of this research.
|
||
|
||
### License
|
||
|
||
[Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0)
|
||
|
||
### Disclaimer
|
||
|
||
This model is intended for general purposes and is available under a permissive Apache License 2.0. Be aware that the model may have biases and/or undesirable outputs. Users deploying systems based on this model are responsible for mitigating risks and complying with applicable AI regulations.
|
||
|
||
### Reference
|
||
```bibtex
|
||
@misc{gplsi-Aitana-7B-S-base,
|
||
author = {Sepúlveda-Torres, Robiert and Baucells, Irene and Estevanell-Valladares, Ernesto L. and Galiano, Santiago and Consuegra-Ayala, Juan Pablo and Miró Maestre, María and Martínez-Murillo, Iván and Grande, Eduardo and Bonora, Mar and Gutierrez, Yoan and Abreu Salas, José Ignacio and Lloret, Elena and Montoyo, Andrés and Muñoz-Guillena and Palomar, Manuel},
|
||
title = {Aitana 7B base: Continually pre-trained on Valencian},
|
||
year = {2026},
|
||
institution = {Language and Information Systems Group (GPLSI) and Centro de Inteligencia Digital (CENID), University of Alicante (UA)},
|
||
howpublished = {\url{https://huggingface.co/gplsi/gplsi/Aitana-2B-S-base}},
|
||
note = {Accessed: 2026-4-8}
|
||
}
|
||
```
|
||
|
||
---
|
||
|
||
**Copyright © 2026 Language and Information Systems Group (GPLSI) and Centro de Inteligencia Digital (CENID),
|
||
University of Alicante (UA).
|
||
Distributed under the Apache License 2.0.**
|