AI-Text-Generation – Tesseract Technolabs

Case Study

project description

In this project, we created an AI model which can generate text based on a given input, and the input will be 1-2 sentences.
Using this model we can generate a story, information, or content related to a given input.
Our model can generate text up to 2-3 paragraphs.
We have used the GPT-J model available for fine-tuning and trained this model on a different dataset collected from different places.
Moreover, for user-friendliness, we have also created a simple UI along with Flask API. So to get a prediction FrontEnd will call the API of Flask and it will return the output.

Technologies used

We created this entire project in Python.
To create the model we have used the PyTorch library of python.
Moreover, we have also used the huggingface transformer library to get the model architecture.

Difficulties we faced

The significant difficulty with this project was the selection of models, which model can be used to generate text with higher coherence and accuracy.
We also faced a problem while training the model, as the model is too big to be trained on a normal CPU.

Solutions

For the model, we tested a few different models available for text generations like GPT-2, GPT-Neo, BERT, Roberta, etc. And selected GPT-J as our final product.
For the training of the model, we selected google colab pro+, as it provides a high-power GPU for training such a model.

NLP

10 months

Share now

gallery

project images

ready to get started?

Receive news, announcement and reports