projecte-aina / FLOR-6.3B-Instructed

FLOR-6.3B Instructed

Table of Contents

Model description

FLOR-6.3B-Instructed is a 6.3B-parameter transformer-based causal language model for Catalan, Spanish, and English, trained on a combined dataset from InstruCat , a Catalan language set of instruction generated automatically from project-aina task orientated dataset, a subset of the Dolly dataset for English, and MENTOR_ES and MENTOR_CA , a Spanish and Catalan sets of instructions commisioned by the BSC Language Technologies Unit. It is the result of a language adaptation technique performed on BLOOM-7.1B , which involves modifying the model's vocabulary and embedding layer, and continuously pre-training the model with 140B tokens in our target languages. Blog post describing the base model: flor-6-3b, a chinchilla compliant model

Intended uses and limitations

The FLOR-6.3B-Instructed model is ready-to-use for some downstream tasks. It can perform text-generation tasks because fine-tuned for specific scenarios, such as summarization, Question Answering, creative writing, etc.

How to use

import torch
from transformers import pipeline

pipe = pipeline("text-generation", model="projecte-aina/FLOR-6.3B-Instructed")

instruction = "Quants habitants té Mataró?"

context = "Mataró és una ciutat de Catalunya, capital de la comarca del Maresme. Situada al litoral mediterrani, a uns 30 km al nord-est de Barcelona, ha estat tradicionalment un centre administratiu de rellevància territorial i un pol de dinamisme econòmic. Compta amb prop de 130.000 habitants, essent actualment la vuitena població del Principat i la tretzena dels Països Catalans. "

# We need to format the prompt and context using ### and \n

def givePrediction(instruction, context, max_new_tokens=50, repetition_penalty=1.2, top_k=50, top_p=0.95, do_sample=True, temperature=0.5)
    text = f"### Instruction\n{{instruction}}\n### Context\n{{context}}\n### Answer\n"
    response = pipe(text.format(instruction=instruction, context=context),temperature=temperature,repetition_penalty=repetition_penalty, max_new_tokens=max_new_tokens,top_k=top_k, top_p=top_p, do_sample=do_sample)[0]["generated_text"]
    answer = response.split("###")[-1][8:-1]
    return answer

answer = givePrediction(instruction, context)

print(answer)
'130 000'

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.

Training

Instruction Data

The training corpus is composed of 140B tokens gathered from web crawlings and public domain data.

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 Departament de la Vicepresidència i de Polítiques Digitals i Territori de la Generalitat de Catalunya within the framework of Projecte AINA .

Disclaimer