Aitana-2B-S-IP-Instruct/README.md

---
license: apache-2.0
language:
  - ca
  - es
  - en
base_model: gplsi/Aitana-2B-S-IP-base
tags:
  - valencian
  - spanish
  - english
  - text-generation
  - instruct
  - intellectual-property
  - alia
  - gplsi
datasets:
  - projecte-aina/InstruCAT
  - projecte-aina/NLUCat
  - projecte-aina/escagleu-64k
  - OpenAssistant/oasst2
  - projecte-aina/oasst1_ca
  - BSC-LT/m-personas
  - projecte-aina/RAG_Multilingual
  - facebook/flores
  - CohereLabs/aya_dataset
  - Unbabel/TowerBlocks-v0.1
  - projecte-aina/MentorES
  - databricks/databricks-dolly-15k
  - yahma/alpaca-cleaned
  - openai/gsm8k
  - Open-Orca/OpenOrca
  - HuggingFaceH4/no_robots
  - LipengCS/Table-GPT
  - projecte-aina/CoQCat
  - allenai/scifact
  - somosnlp/LingComp_QA
  - somosnlp/instruct-legal-refugiados-es
  - somosnlp-hackathon-2025/gastronomia-hispana-dpo
  - gplsi/boua_parallel
library_name: transformers
pipeline_tag: text-generation
---
# Aitana-2B-S-IP-Instruct

**Aitana-2B-S-IP-Instruct** is an instruction-tuned generative language model from the **Aitana family**, developed by the [GPLSI (Language and Information Systems Group)](https://gplsi.dlsi.ua.es/) at the University of Alicante. Built on Aitana-2B-S-IP-Instruct, this model has been fine-tuned to follow instructions across Valencian, Spanish, and English, with a specialized focus on intellectual property domain tasks.

## Table of Contents

- [Model Description](#model-description)
- [Training Data](#training-data)
- [Intended Uses](#intended-uses)
- [How to Use](#how-to-use)
- [Evaluation](#evaluation)
- [Additional Information](#additional-information)

## Model Description

| Property | Value |
|----------|-------|
| **Base Model** | [gplsi/Aitana-2B-S-IP-Instruct](https://huggingface.co/gplsi/Aitana-2B-S-IP-Instruct) |
| **Architecture** | Transformer decoder-only |
| **Parameters** | ~2.25B |
| **Languages** | Valencian, Spanish, English |
| **License** | Apache 2.0 |

Aitana-2B-S-IP-Instruct is an instruction-tuned variant of Aitana-2B-S-IP-Instruct, fine-tuned on multilingual instruction data with emphasis on intellectual property applications.

## Training Data

This model was instruction fine-tuned using the following data:

| Dataset ID | Name | Languages | Source |
|------------|------|-----------|--------|
| ins1 | InstruCAT | CA | [projecte-aina/InstruCAT](https://huggingface.co/datasets/projecte-aina/InstruCAT) |
| ins2 | NLUCat | CA | [projecte-aina/NLUCat](https://huggingface.co/datasets/projecte-aina/NLUCat) |
| ins3 | Escagleu 64K | CA | [projecte-aina/escagleu-64k](https://huggingface.co/datasets/projecte-aina/escagleu-64k) |
| ins4 | OpenAssistant2 (OASST2) | CA, EN, ES, VA | [OpenAssistant/oasst2](https://huggingface.co/datasets/OpenAssistant/oasst2) |
| ins5 | OpenAssistant1 (OASST1) | CA, VA | [projecte-aina/oasst1_ca](https://huggingface.co/datasets/projecte-aina/oasst1_ca) |
| ins6 | M-Personas | CA, EN, ES, VA | [BSC-LT/m-personas](https://huggingface.co/datasets/BSC-LT/m-personas) |
| ins7 | RAG Multilingual | CA, EN, ES | [projecte-aina/RAG_Multilingual](https://huggingface.co/datasets/projecte-aina/RAG_Multilingual) |
| ins8 | FLORES | CA, EN, ES | [facebook/flores](https://huggingface.co/datasets/facebook/flores) |
| ins9 | Aya Dataset | EN, ES, VA | [CohereLabs/aya_dataset](https://huggingface.co/datasets/CohereLabs/aya_dataset) |
| ins10 | TowerBlocks | EN, ES | [Unbabel/TowerBlocks-v0.1](https://huggingface.co/datasets/Unbabel/TowerBlocks-v0.1) |
| ins11 | Mentor / Mentores | CA, ES, VA | [projecte-aina/MentorES](https://huggingface.co/datasets/projecte-aina/MentorES) / [projecte-aina/MentorCA](https://huggingface.co/datasets/projecte-aina/MentorCA) |
| ins12 | Dolly / Dolly 3K | CA, EN, VA | [databricks/databricks-dolly-15k](https://huggingface.co/datasets/databricks/databricks-dolly-15k) / [projecte-aina/dolly3k_ca](https://huggingface.co/datasets/projecte-aina/dolly3k_ca) |
| ins13 | Alpaca | EN, VA | [yahma/alpaca-cleaned](https://huggingface.co/datasets/yahma/alpaca-cleaned) |
| ins14 | GSM8K | EN, VA | [openai/gsm8k](https://huggingface.co/datasets/openai/gsm8k) |
| ins15 | OpenOrca | EN | [Open-Orca/OpenOrca](https://huggingface.co/datasets/Open-Orca/OpenOrca) |
| ins16 | No Robots | EN | [HuggingFaceH4/no_robots](https://huggingface.co/datasets/HuggingFaceH4/no_robots) |
| ins17 | TableGPT | EN | [LipengCS/Table-GPT](https://huggingface.co/datasets/LipengCS/Table-GPT) |
| ins18 | CoQCA / CoQCat | CA, VA | [projecte-aina/CoQCat](https://huggingface.co/datasets/projecte-aina/CoQCat) |
| ins19 | SciFact | EN, VA | [allenai/scifact](https://huggingface.co/datasets/allenai/scifact) |
| ins20 | LingComp QA | ES, VA | [somosnlp/LingComp_QA](https://huggingface.co/datasets/somosnlp/LingComp_QA) |
| ins21 | Instruct Legal Refugiados | ES, VA | [somosnlp/instruct-legal-refugiados-es](https://huggingface.co/datasets/somosnlp/instruct-legal-refugiados-es) |
| ins22 | Gastronomia Hispana | ES, VA | [somosnlp-hackathon-2025/gastronomia-hispana-dpo](https://huggingface.co/datasets/somosnlp-hackathon-2025/gastronomia-hispana-dpo) |
| ins23 | TurismInstructionsGPLSI | VA | — |
| ins24 | Amic-Paralelo | VA | — |
| ins25 | BOUA | VA | [gplsi/boua_parallel](https://huggingface.co/datasets/gplsi/boua_parallel) |
| ins26 | DOGV Parallel | VA | — |
| ins27 | UJI VA-EN Parallel | VA | — |
| ins28 | UJI VA-ES Parallel | VA | — |

## Intended Uses

This model can be used for:

- **Instruction following** in Valencian, Spanish, and English
- **Intellectual property domain** applications
- **Chat and conversational applications** requiring multilingual support
- **Text generation** with task-specific prompting

## How to Use

### Transformers

```python
import torch
from transformers import pipeline, AutoTokenizer

model_id = "gplsi/Aitana-2B-S-IP-Instruct-IP"
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 = "Explica què és la propietat intel·lectual i quins drets atorga."
result = generator(text, do_sample=True, top_k=10, max_new_tokens=100)
print(result[0]['generated_text'])
# Spanish example
text = "Describe los principales tipos de propiedad intelectual y su marco legal."
result = generator(text, do_sample=True, top_k=10, max_new_tokens=100)
print(result[0]['generated_text'])
# English example
text = "Explain the concept of intellectual property and its importance in innovation."
result = generator(text, do_sample=True, top_k=10, max_new_tokens=100)
print(result[0]['generated_text'])
```

## Evaluation

In the following tables, we present the results obtained with different benchmarks from [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) in comparison with [Salamandra-2B-Instruct](https://huggingface.co/BSC-LT/Salamandra-2B-Instruct).

### Normalized score per language

| Language | Salamandra-2B-Instruct | Aitana-2B-S-IP-Instruct |
|----------|------------------------|-----------------------------|
| Spanish | 0.079 | **0.112** |
| Catalan | **0.202** | 0.182 |
| English | **0.178** | 0.167 |
| Valencian | **0.507** | 0.489 |
| **Average** | **0.242** | 0.237 |

### Valencian

#### Classification Benchmarks

| Dataset | Lang. | Task | Metric | Salamandra-2B-Instruct | Aitana-2B-S-IP-Instruct |
|---------|--------|------|--------|------------------------|-----------------------------|
| XNLI | va | Natural Language Inference | acc | **0.520** | 0.501 |

#### Generation Benchmarks

| Dataset | Lang. | Task | Metric | Salamandra-2B-Instruct | Aitana-2B-S-IP-Instruct |
|---------|--------|------|--------|------------------------|-----------------------------|
| Cocoteros | va | Reading Comprehension | bleu | 2.796 | **3.204** |
| Phrases ca-va | va-ca | Translation - Adaptation | bleu | 58.425 | **58.694** |
| Phrases va-ca | va-ca | Translation - Adaptation | bleu | **70.660** | 56.706 |
| Phrases va-es | va-es | Translation | bleu | **65.427** | 53.129 |
| Phrases es-va | es-va | Translation | bleu | **45.688** | 43.098 |
| Truthfulqa_va | va | Truthfulness | bleu_acc | **0.409** | 0.381 |

### Catalan

#### Classification Benchmarks

| Dataset | Lang. | Task | Metric | Salamandra-2B-Instruct | Aitana-2B-S-IP-Instruct |
|---------|--------|------|--------|------------------------|-----------------------------|
| Belebele Cat_latn | ca | Reading Comprehension | acc | **0.287** | 0.253 |
| COPA | ca | Commonsense Reasoning | acc | **0.708** | 0.706 |
| XStoryCloze | ca | Commonsense Reasoning | acc | **0.616** | **0.616** |
| OpenBookQA | ca | Question Answering | acc | **0.296** | 0.270 |
| PAWS | ca | Paraphrasing | acc | 0.602 | **0.603** |
| PiQA | ca | Question Answering | acc | 0.638 | **0.643** |
| SiQA | ca | Question Answering | acc | **0.422** | 0.421 |
| ARC Easy | ca | Question Answering | acc | **0.516** | 0.501 |
| ARC Challenge | ca | Question Answering | acc | 0.298 | **0.299** |
| XNLI | ca | Natural Language Inference | acc | 0.513 | **0.517** |
| Teca | ca | Natural Language Inference | acc | 0.486 | **0.494** |
| WNLI | ca | Natural Language Inference | acc | **0.563** | 0.437 |
| Catcola | ca | Linguistic Acceptability | acc | 0.492 | **0.718** |
| Catcola | ca | Linguistic Acceptability | mcc | **0.097** | -0.034 |
| Catalanqa | ca | Question Answering | F1 | **0.516** | 0.397 |
| Mgsm direct | ca | Math | exact match | 0.000 | 0.000 |
| Catalanqa | ca | Question Answering | exact match | **0.182** | 0.049 |
| Xquad | ca | Question Answering | exact match | **0.103** | 0.055 |
| Xquad | ca | Question Answering | F1 | **0.394** | 0.312 |

#### Generation Benchmarks

| Dataset | Lang. | Task | Metric | Salamandra-2B-Instruct | Aitana-2B-S-IP-Instruct |
|---------|--------|------|--------|------------------------|-----------------------------|
| Cabreu abstractive | ca | Summarization | bleu | 7.610 | **8.516** |
| Cabreu extractive | ca | Summarization | bleu | **38.002** | 31.230 |
| Cabreu extreme | ca | Summarization | bleu | 2.733 | **3.070** |

### Spanish

#### Classification Benchmarks

| Dataset | Lang. | Task | Metric | Salamandra-2B-Instruct | Aitana-2B-S-IP-Instruct |
|---------|--------|------|--------|------------------------|-----------------------------|
| Belebele | es | Reading Comprehension | acc | **0.268** | **0.268** |
| PAWS | es | Paraphrasing | acc | 0.566 | **0.623** |
| XNLI | es | Natural Language Inference | acc | **0.463** | 0.442 |
| WNLI | es | Natural Language Inference | acc | **0.479** | 0.451 |
| XStoryCloze | es | Commonsense Reasoning | acc | **0.617** | 0.614 |
| Escola | es | Linguistic Acceptability | acc | 0.293 | **0.662** |
| Escola | es | Linguistic Acceptability | mcc | **0.020** | 0.000 |
| OpenbookQA | es | Question Answering | acc | 0.286 | **0.296** |
| MGSM Direct | es | Math | exact match | 0.020 | **0.060** |
| XQUAD | es | Question Answering | exact match | **0.066** | 0.035 |
| XQUAD | es | Question Answering | F1 | **0.355** | 0.292 |

#### Generation Benchmarks

| Dataset | Lang. | Task | Metric | Salamandra-2B-Instruct | Aitana-2B-S-IP-Instruct |
|---------|--------|------|--------|------------------------|-----------------------------|
| Cocoteros | es | Reading Comprehension | bleu | **3.308** | 2.755 |
| XLSum | es | Summarization | bleu | **1.695** | 1.474 |

### English

#### Classification Benchmarks

| Dataset | Lang. | Task | Metric | Salamandra-2B-Instruct | Aitana-2B-S-IP-Instruct |
|---------|--------|------|--------|------------------------|-----------------------------|
| Arc Challenge | en | Question Answering | acc | **0.354** | 0.348 |
| Arc Easy | en | Question Answering | acc | 0.681 | **0.693** |
| Belebele | en | Reading Comprehension | acc | 0.260 | **0.267** |
| PAWS | en | Paraphrasing | acc | 0.597 | **0.602** |
| XNLI | en | Natural Language Inference | acc | 0.512 | **0.547** |
| XStoryCloze | en | Commonsense Reasoning | acc | **0.662** | 0.655 |
| OpenBookQA | en | Question Answering | acc | 0.298 | **0.308** |
| PiQA | en | Question Answering | acc | 0.715 | **0.721** |
| Social iqa | en | Question Answering | acc | **0.453** | 0.419 |
| WNLI | en | Natural Language Inference | acc | **0.535** | 0.437 |
| MGSM Direct | en | Math | exact match | 0.008 | **0.080** |
| TriviaQA | en | Question Answering | exact match | 0.076 | **0.095** |
| CoLA | en | Linguistic Acceptability | mcc | **0.055** | -0.008 |

### Judge Evaluation

The model was also evaluated using an LLM-as-judge approach across different task categories. The scores below represent the average rating (1-5 scale, 5 being best) and standard deviation for each task category, comparing against Salamandra-2B-Instruct.

| Task Category | Salamandra-2B-Instruct | Aitana-2B-S-IP-Instruct |
|---------------|------------------------|-----------------------------|
| CommonSense reasoning | **2.277 / 1.151** | 1.891 / 0.934 |
| Maths | 1.060 / 0.124 | **1.075 / 0.151** |
| Paraphrasing | 3.518 / 1.308 | **3.536 / 1.348** |
| Reading comprehension | **2.966 / 1.111** | 2.599 / 1.331 |
| Summarization | **2.217 / 1.068** | 1.827 / 0.822 |
| Translation | **3.557 / 0.760** | 3.502 / 1.031 |
| **Overall Avg** | **2.599 / 0.920** | 2.405 / 0.936 |

## Additional Information

### Author

The model has been developed by the **Language and [Information Systems Group (GPLSI)](https://gplsi.dlsi.ua.es/)** and the **[Centro de Inteligencia Digital (CENID)](https://cenid.es)**, both part of the **[University of Alicante (UA)](https://www.ua.es/es/)**, as part of their ongoing research in **Natural Language Processing (NLP)**.


### 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 Technologies Laboratory at Barcelona Supercomputing Center](https://www.bsc.es/es/discover-bsc/organisation/research-structure/language-technologies-laboratory)
- [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-2B-S-IP-Instruct,
  author       = {Martínez-Murillo, Iván and Sepúlveda-Torres, Robiert and Grande, Eduardo and Galiano, Santiago and Estevanell-Valladares, Ernesto L. and Consuegra-Ayala, Juan Pablo and Miró Maestre, María and Canal-Esteve, Miquel and Bonora, Mar and Gutierrez, Yoan and Abreu Salas, José Ignacio and Lloret, Elena and Montoyo, Andrés and Muñoz-Guillena, Rafael and Palomar, Manuel},
  title        = {Aitana 2B Instruct IP: Instruction-tuned model for intellectual property applications in Valencian, Spanish and English},
  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/Aitana-2B-S-IP-Instruct}},
  note         = {Accessed: 2026-05-21}
}
```

---

**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.**