Training a transformer pipeline

Hello guys, I am trying to generate a real estate description based on some features. I tried using the OpenAI GPT-3 solution but it turned out to be so expensive. I found that I can use the GPT-Neo alternative or the t5 model. I am currently trying to work with the transformers Pipelines Pipelines — transformers 4.7.0 documentation
The output of summarization isn’t that bad. For example when I input :
“convert the following features to a real estate listing description: \n apartment for sale in La Marsa \n 2 bedrooms and one living room \n no pool \n nice view on park \n near to transport \n near to Bank”
this is the output I had
“apartment for sale in la Marsa, california . 2 bedrooms and 1 living room no pool . close to transport and to the bank of the san antonio . no pool and no pool in front of the apartment . the following short hand to a real estate listing description: a condo for rent in la marsa with a pool and a gym . for more information about this property, visit
(La Marsa is a city in Tunisia which makes the generated paragraph irrelevant)
Is there a way to train the model ? If you can help me with any ideas please do. I’m new to this. Thanks