Turkish big language model from İş Bank: İşGPT

İş Bankası developed the Turkish major language model İşGPT. Developed by İşbank Artificial Intelligence Department İşGPTIt can be actively used by all employees in various application areas such as legislation, foreign trade, performance management and content structuring.

İşGPT, which has advanced artificial intelligence functions such as text summarization, question answering through content, classification, Named Entity Recognition (NER) and translation, answers 91 percent of the questions asked quickly and accurately. With all these features, the bank aims to accelerate customer service processes and further increase the productivity of its employees.

According to what was shared, more than 1.5 million documents in total were processed for training the model, using a wide set of documents compiled from various sectors and fields during the FineTune process of the model. These documents were selected from sources that cover various fields such as finance, health, law, education and technology and contain up-to-date information.

The total number of tokens used during the training process was calculated as 60 billion. This number of tokens was chosen so that the model can adapt to a wide range of languages ​​and complex language structures.

The training of the model was completed in a total of 1200 hours. During this time, detailed hyperparameter adjustments were made to optimize the model’s performance and achieve the highest accuracy.

İşbank Deputy General Manager Sabri Gökmenler, “İşGPT provides speed and accuracy in banking transactions, while taking customer and employee experience to higher levels. In this project, utmost care is taken to ensure that our data does not leave the country.” said.