Latxa Llama 3.1 70B Instruct

🚀 HiTZ/Latxa-Llama-3.1-70B-Instruct

We introduce Latxa 3.1 70B Instruct, an instructed version of Latxa, which outperforms Llama-3.1-Instruct on Basque benchmarks and shows great potential in chat conversations.

🚀 Quick Start

Use the code below to get started with the model.

from transformers import pipeline

pipe = pipeline('text-generation', model='HiTZ/Latxa-Llama-3.1-70B-Instruct')

messages = [
    {'role': 'user', 'content': 'Kaixo!'},
]

pipe(messages)

>>
[
  {
    'generated_text': [
      {'role': 'user', 'content': 'Kaixo!'},
      {'role': 'assistant', 'content': 'Kaixo! Zer moduz? Zer behar edo galdetu nahi duzu?'}
    ]
  }
]

✨ Features

• High Performance on Basque: Our preliminary experimentation shows that Latxa 3.1 70B Instruct outperforms Llama-3.1-Instruct by a large margin on Basque standard benchmarks, especially in chat conversations.
• Good Ranking in Public Evaluation: In a public arena-based evaluation, Latxa ranked 3rd, just behind Claude and GPT-4 and above all the other same-size competitors.

📦 Installation

The installation details are related to the transformers library. You can install it using the following command:

pip install transformers

💻 Usage Examples

Basic Usage

from transformers import pipeline

pipe = pipeline('text-generation', model='HiTZ/Latxa-Llama-3.1-70B-Instruct')

messages = [
    {'role': 'user', 'content': 'Kaixo!'},
]

result = pipe(messages)
print(result)

📚 Documentation

Model Details

Model Description

Latxa is a family of Large Language Models (LLM) based on Meta’s LLaMA models. Current LLMs perform incredibly well for high-resource languages like English. However, for Basque and other low-resource languages, their performance is close to a random guesser. These limitations widen the gap between high- and low-resource languages in digital development. We present Latxa to overcome these limitations and promote the development of LLM-based technology and research for the Basque language. Latxa models follow the same architecture as their original counterparts and were further trained in Latxa Corpus v1.1, a high-quality Basque corpora.

Uses

Latxa models are intended to be used with Basque data; for any other language, the performance is not guaranteed. Similar to the original, Latxa inherits the Llama-3.1 License, which allows for commercial and research use.

Direct Use

Latxa Instruct models are trained to follow instructions or work as chat assistants.

Out-of-Scope Use

The model is not intended for malicious activities, such as harming others or violating human rights. Any downstream application must comply with current laws and regulations. Irresponsible usage in production environments without proper risk assessment and mitigation is also discouraged.

Bias, Risks, and Limitations

In an effort to alleviate potentially disturbing or harmful content, Latxa has been trained on carefully selected and processed data, mainly from local media, national/regional newspapers, encyclopedias, and blogs (see Latxa Corpus v1.1). However, the model is based on Llama 3.1 models and may potentially carry the same bias, risk, and limitations. Please refer to Llama’s Ethical Considerations and Limitations for further information.

Training Details

⚠️ Important Note

Further training details will be released with the corresponding research paper in the near future.

Evaluation

We evaluated the models in 5-shot settings on multiple-choice tasks, using the Basque partitions of each dataset. The arena results will be released in the future.

Testing Data, Factors & Metrics

Testing Data

• Belebele (Bandarkar et al.): Belebele is a multiple-choice machine reading comprehension (MRC) dataset covering 122 language variants. We evaluated the model in a 5-shot manner.
  • Data card: https://huggingface.co/datasets/facebook/belebele
• X-StoryCloze (Lin et al.): XStoryCloze is a professionally translated version of the English StoryCloze dataset into 10 non-English languages. Story Cloze is a commonsense reasoning dataset that requires choosing the correct ending for a four-sentence story. We evaluated the model in a 5-shot manner.
  • Data card: https://huggingface.co/datasets/juletxara/xstory_cloze
• EusProficiency (Etxaniz et al., 2024): EusProficiency contains 5,169 exercises on different topics from past EGA exams, the official C1-level certificate of proficiency in Basque.
  • Data card: https://huggingface.co/datasets/HiTZ/EusProficiency
• EusReading (Etxaniz et al., 2024): EusReading consists of 352 reading comprehension exercises (irakurmena) from the same set of past EGA exams.
  • Data card: https://huggingface.co/datasets/HiTZ/EusReading
• EusTrivia (Etxaniz et al., 2024): EusTrivia includes 1,715 trivia questions from multiple online sources. 56.3% of the questions are at the elementary level (grades 3 - 6), while the rest are considered challenging.
  • Data card: https://huggingface.co/datasets/HiTZ/EusTrivia
• EusExams (Etxaniz et al., 2024): EusExams is a collection of tests designed to prepare individuals for Public Service examinations conducted by several Basque institutions, including the public health system Osakidetza, the Basque Government, the City Councils of Bilbao and Gasteiz, and the University of the Basque Country (UPV/EHU).
  • Data card: https://huggingface.co/datasets/HiTZ/EusExams

Metrics

We use Accuracy since the tasks are framed as Multiple Choice questions.

Results

Environmental Impact

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

• Hardware Type: HPC Cluster, 4 x A100 64Gb nodes x64
• Hours used (total GPU hours): 16005.12h
• Cloud Provider: CINECA HPC
• Compute Region: Italy
• Carbon Emitted: 1901.41kg CO2 eq

Acknowledgements

This work has been partially supported by the Basque Government (IKER-GAITU project). It has also been partially supported by the Ministerio para la Transformación Digital y de la Función Pública - Funded by EU – NextGenerationEU within the framework of the project with reference 2022/TL22/00215335. The models were trained on the Leonardo supercomputer at CINECA under the EuroHPC Joint Undertaking, project EHPC-EXT-2023E01-013.

Citation

Coming soon.

Meanwhile, you can reference:

@misc{etxaniz2024latxa,
    title={{L}atxa: An Open Language Model and Evaluation Suite for {B}asque},
    author={Julen Etxaniz and Oscar Sainz and Naiara Perez and Itziar Aldabe and German Rigau and Eneko Agirre and Aitor Ormazabal and Mikel Artetxe and Aitor Soroa},
    year={2024},
    eprint={2403.20266},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}