Carballo-bloom-1.3B

Table of Contents

• Carballo-bloom-1.3B
  • Table of Contents
  • Model description
  • Intended uses and limitations
  • How to use
  • Training
    • Tools
    • Language adaptation and training
    • Training data
    • Training hyperparameters
    • Framework
  • Evaluation
  • Additional information
    • Contact
    • Copyright
    • License
    • Funding
  • Citation information

Model description

Carballo-bloom-1.3B is a 1.3B-parameter transformer-based causal language model for Galician. It is the result of a continual pretraining of FLOR-1.3B (developed by AINA Project and based in BLOOM-1.7B) with the galician corpus CorpusNos.

Intended uses and limitations

The Carballo-bloom-1.3B model is ready-to-use only for causal language modeling. It can perform text-generation tasks and be fine-tuned for specific scenarios.

How to use

import torch
from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM

input_text = "Hoxe fai un bo día. O sol  "

model_id  = "proxectonos/Carballo-bloom-1.3B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)
generator = pipeline(
    "text-generation",
    model=model,
    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']}")

Training

Tools

It was trained using HuggingFace Transformers and Pytorch, using the Causal Modeling Language script

Language adaptation and training

The language adaptation technique used to train Carballo-bloom-1.3B is based in the used to train FLOR-1.3B, which is explained by their authors in this Medium Post. In summary, we proceeded as follows:

1. We trained our own BPE tokenizer for galician and replaced the original FLOR-1.3B tokenizer and vocabulary with it.
2. The embeddings corresponding to tokens that are present in both the original and the target vocabulary (matching tokens) were used for initialization.
3. The embeddings from tokens not present in Carballo-bloom-1.3B's original vocabulary were initialized as the average of all embeddings.
4. The model was initialized with the weights from FLOR-1.3B and with our adapted tokenizer (step 1) and embeddings (steps 2-3).
5. The model was then trained on a galician corpus.

Training data

CorpusNÓS is a massive Galician corpus made up of 2.1B words primarily devised for training large language models. The corpus sources are varied and represent a relatively wide range of genres.

The corpus is structured as follows:

Training hyperparameters

• seed: 42
• num_devices: 1
• train_batch_size: 2
• eval_batch_size: 2
• gradient_acummulation: 4
• optimizer: AdamW
• betas: (0.9,0.999)
• epsilon: 1e-08
• weight_decay_rate: 0.1
• scheduler: "Linear"
• learning_rate: 5e-05
• num_epochs: 1.2

Framework

The traininf was conducted in the Galicia Supercomputing Center (CESGA), using 1 node with 5 GPUs NVIDIA A100.

Evaluation

Additional information

Contact

For further information, please send an email to proxecto.nos@usc.gal

Funding

This model was development within the Nós Project, funded by the Ministerio para la Transformación Digital y de la Función Pública - Funded by EU – NextGenerationEU within the framework of the project ILENIA with reference 2022/TL22/00215336.

How to cite this work

if you use this model, please cite this article:

Gamallo, Pablo, Pablo Rodríguez Fernández, Iria de Dios Flores, Susana Sotelo, Silvia Paniagua, José Ramom Pichel, Daniel Bardanca, Marcos Garcia (2024) "Open Generative Large Language Models for Galician", Procesamiento del Lenguaje Natural, 73, pp. 259-270. ISSN: 1135-5948.