Part of the TPC Seminar Series

---

Abstract:

Data generation for supervised fine-tuning of Large LanguageModels (LLMs) is a significant challenge due to its severe computational and time requirements. To our knowledge, no publicly available tool exists that efficiently generates a scientific corpus for supervised training or fine-tuning of LLMs. This paper presents a programmable framework for scraping, parsing, and formatting scientific literature to prepare a corpus in question-answer format to train or fine-tune LLMs using PDFs. Finally, we apply this tool to curate approximately 10𝐾+ PDFs culminating into 415𝐾+ questions and their answers within a month, along with appropriate keywordsto describe them based on HPC performance optimization.

Biography:

Avaneesh Ramesh is a senior in the IB Program at Westwood High School in Austin, Texas. He has been working with Dr. Tanzima Islam, an assistant professor in the Department of Computer Science at Texas State University, in her Per4ML Laboratory since 2021. At Per4ML, he has researched at the intersections between Machine Learning, Visualization, and High-Performance-Computing, contributing to research papers, extended abstracts, and poster presentations. Outside of Per4ML, Avaneesh has continued to further his knowledge on computer science, most notably with his 4000 word Extended Essay submission to IB on comparing the efficacies of various neural network technologies in bone break recognition. In addition to furthering his own knowledge, Avaneesh has taken teaching positions to help young students understand computer science, with his experiences as a STEM tutor at LearnLee, a cofounder of the CodeClub Austin, and a contributor to a computer science blog at the Austin Public Library. In his free time, Avaneesh enjoys spending time with family and friends, playing the clarinet in his high school’s Wind Ensemble and Marching Band, and performing classical guitar pieces.

Additional Zoom Information:

Meeting ID: 161 278 3531
Passcode: 929445

One tap mobile
+16692545252,,1612783531#<tel:+16692545252,1612783531>,,,,*929445# US (San Jose)
+16469641167,,1612783531#<tel:+16469641167,1612783531>,,,,*929445# US (US Spanish Line)

Dial by your location
+1 669 254 5252<tel:+16692545252> US (San Jose)
+1 646 964 1167<tel:+16469641167> US (US Spanish Line)
+1 646 828 7666<tel:+16468287666> US (New York)
+1 669 216 1590<tel:+16692161590> US (San Jose)
+1 415 449 4000<tel:+14154494000> US (US Spanish Line)
+1 551 285 1373<tel:+15512851373> US (New Jersey)