Abstract: This paper presents a transformer machine-learning model trained with a large dataset of chord sequences. The dataset includes several styles, such as jazz, rock, pop, blues, or music for cinema, among others. We investigated three modeling strategies: 1) We started the tokenization method by treating all different chords as unique elements, which resulted in a vocabulary of 5202 independent chords as tokens. 2) We expressed the chords as a tuple describing root, nature (e.g., major, minor, diminished, major seventh), and extensions (e.g., additions, alterations), which produces a vocabulary of 59 tokens. 3) We extended the second model by complementing the transformer model with chord information containing eight MIDI notes added to the positional embeddings. We analyze sequences generated by comparing them with the training dataset using trigram analysis, which reveals common chord progressions and source duplications. Secondly, we compared the generated sequences from a musical perspective, rating their plausibility concerning the training data. The third strategy reported lower validation loss and more musical consistency in the suggested progressions.