PersianTechSupportAssistant

This project involves the development of a tech support assistant in Persian language using large language models (LLMs). The project consists of three main parts: initial data gathering, data generation, and model fine-tuning. In the first part, a corpus of texts related to technology and gadgets like mobiles and laptops was collected from Digikala Mag and Digiato Persian blogs. In the second part, ChatGPT 3.5 Turbo via OpenAI API was used to generate question-answer pairs from the crawled corpus. GPT2 was trained on the corpus and fine-tuned on the question-answer data in the third part. Additionally, a Persian Llama named ‘persian_llama_7b’ was fine-tuned using Hugging Face tools on the question-answer data. Parameter-efficient fine-tuning methods like QLORA were used to enable Colab and Kaggle resources to solve the task. The resulting model can be used as a tech support assistant in Persian language. The paper provides insights into the challenges faced during the project and how they were overcome. Future work that can be done in this field is also discussed.