# Aligning a medium-size GPT model in English to a small closed domain in Spanish

arXiv: https://arxiv.org/abs/2303.17649 · HTML: https://mlsanity.com/paper/2303.17649/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Reward modeling and RLHF, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 6 of its 13 references that are in the map.
- First posted on arXiv 30 Mar 2023.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Reward modeling and RLHF](https://mlsanity.com/alignment/reward-modeling-and-rlhf/)

## Builds on (most cited first, 6 of 6)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Fine-Tuning Language Models from Human Preferences](https://mlsanity.com/paper/1909.08593/) (2019)
- [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://mlsanity.com/paper/1911.00536/) (2019)
- [Smaaprat: DialoGPT for Natural Language Generation of Swedish Dialogue by Transfer Learning](https://mlsanity.com/paper/2110.06273/) (2021)

## Abstract (arXiv)

In this paper, we propose a methodology to align a medium-sized GPT model, originally trained in English for an open domain, to a small closed domain in Spanish. The application for which the model is finely tuned is the question answering task. To achieve this we also needed to train and implement another neural network (which we called the reward model) that could score and determine whether an answer is appropriate for a given question. This component served to improve the decoding and generation of the answers of the system. Numerical metrics such as BLEU and perplexity were used to evaluate the model, and human judgment was also used to compare the decoding technique with others. Finally, the results favored the proposed method, and it was determined that it is feasible to use a reward model to align the generation of responses.