Show simple item record

dc.contributor.advisorShrestha, Raju
dc.contributor.authorChatterjee, Shiladitya
dc.date.accessioned2024-06-11T11:16:42Z
dc.date.available2024-06-11T11:16:42Z
dc.date.issued2023
dc.identifier.urihttps://hdl.handle.net/11250/3133516
dc.description.abstractThe car industry at the Indian subcontinent has been a booming industry over the decade. As new and new cars have arrived at the Indian market it becomes extremely difficult for a consumer to keep track of the cars launched and their details. The paper outlines the attempt to provide a solution to the end-user/customer in the form of generative model containing all the relevant information of all the hatchback cars available in the Indian automobile market. The distributed data was collated and carefully curated to create a unique dataset containing details of 23 cars which are now present in the Indian market. The dataset used for this experiment has been curated from data taken from over 10 websites primarily containing the different data of each car. A generative model based purely on cars has been developed and it is called GenerativeAutoAi. The power of transfer learning has been leveraged to train the language model GPT-2. The GPT-2 had a BLEU score of 0.78. The new innovative approach of fine tuning was used to tune a large language model Falcon7B. The Falcon 7B model was trained with the help of QLora. This model has just been launched in the hugging face family for commercial use hence a lot of research has not gone into how it can be further leveraged. It has primarily been used to showcase the use of the created data and with just 650+ pairs of Question and Answers it has showed remarkable fluency.en_US
dc.language.isoengen_US
dc.publisherOslomet - storbyuniversiteteten_US
dc.titleGenerativeAutoAi - A generative model for carsen_US
dc.typeMaster thesisen_US
dc.description.versionpublishedVersionen_US


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record