projecte-aina / aguila-7b

Ǎguila-7B

Table of Contents

Model description

Ǎguila-7B is a transformer-based causal language model for Catalan, Spanish, and English. It is based on the Falcon-7B model and has been trained on a 26B token trilingual corpus collected from publicly available corpora and crawlers.

More information available in the following post from Medium.com: Introducing Ǎguila, a new open-source LLM for Spanish and Catalan ( https://medium.com/@mpamies247/introducing-a%CC%8Cguila-a-new-open-source-llm-for-spanish-and-catalan-ee1ebc70bc79 )

Intended uses and limitations

The Ǎguila-7B model is ready-to-use only for causal language modeling to perform text-generation tasks. However, it is intended to be fine-tuned for downstream tasks.

How to use

Here is how to use this model:

import torch
from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM

input_text = "El mercat del barri és fantàstic, hi pots trobar"

model_id  = "projecte-aina/aguila-7b"
tokenizer = AutoTokenizer.from_pretrained(model_id)
generator = pipeline(
    "text-generation",
    model=model_id,
    tokenizer=tokenizer,
    torch_dtype=torch.bfloat16,
    trust_remote_code=True,
    device_map="auto",
)
generation = generator(
    input_text,
    do_sample=True,
    top_k=10,
    eos_token_id=tokenizer.eos_token_id,
)

print(f"Result: {generation[0]['generated_text']}")

Limitations and bias

At the time of submission, no measures have been taken to estimate the bias and toxicity embedded in the model. However, we are well aware that our models may be biased since the corpora have been collected using crawling techniques on multiple web sources. We intend to conduct research in these areas in the future, and if completed, this model card will be updated.

Language adaptation

We adapted the original Falcon-7B model to Spanish and Catalan by swapping the tokenizer and adjusting the embedding layer.

The adaptation procedure is explained in this blog post .

Training

Training data

The training corpus consists of 26B tokens of several corpora gathered from web crawlings and public domain data.

The dataset has the following language distribution:

Note: A small amount of English data was kept to avoid catastrophic forgetting.

Training procedure

The training corpus has been tokenized using a byte version of Byte-Pair Encoding (BPE) with a vocabulary size of 50,257 tokens. After training a new tokenizer and adapting falcon-7b 's embedding layer, the model was further pre-trained in three target languages: Catalan, Spanish and English.

The training lasted a total of 320 hours on 8 NVIDIA H100 GPUs with 80GB RAM.

Training hyperparameters

• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• train_batch_size: 1
• eval_batch_size: 1
• total_train_batch_size: 8
• total_eval_batch_size: 8
• optimizer: Adam
• betas: (0.9,0.999)
• epsilon: 1e-08
• learning_rate: 5e-05
• lr_scheduler_type: linear
• num_epochs: 1.0

Framework versions

• Pytorch 2.0.0
• Transformers 4.30.2
• Datasets 2.13.1
• Tokenizers 0.13.3

Additional information

Author

The Language Technologies Unit from Barcelona Supercomputing Center.

Contact

For further information, please send an email to [email protected] .

Copyright

Copyright(c) 2023 by Language Technologies Unit, Barcelona Supercomputing Center.

License

Apache License, Version 2.0

Funding

This work was funded by:

• The Departament de la Vicepresidència i de Polítiques Digitals i Territori de la Generalitat de Catalunya within the framework of Projecte AINA .
• The Spanish State Secretariat for Digitalization and Artificial Intelligence within the framework of the Plan de Impulso de las Tecnologías del Lenguaje .

Disclaimer