2024-07-11 –, For Loop (3.2)
Large Language models(LLMs) such as ChatGPT, Bard, Midjourney, Gemini, and many others are buzzwords in the present era of Artificial Intelligence. In this talk, I will address the impact of large language models on the environment and present small language models as the solution.
Large Language models(LLMs) such as ChatGPT, Bard, Midjourney, Gemini, and many others are buzzwords in the present era of Artificial Intelligence. The technologies such as Large Language Models are like a double-edged sword. On one hand, they offer lots of benefits and create new opportunities and on the other hand, they raise many concerns. One of the concerns, that is least discussed is the impact of large language models on the environment. The large language models are trained on billions and trillions of parameters and they require huge compute resources that contribute to carbon footprint. One of the possible solutions to reduce the carbon footprint and environmental impact of LLMs is to consider using Small Language Models and Modular Language Models. In many cases, small language models offer better solutions than large language models. This talk begins with an introduction to Large Language models and their impact on the environment and how they contribute to carbon footprint. Next, I will introduce Small Language models and how they will help in reducing the carbon footprint while offering a better solution. I will demonstrate a sample small language model application using Julia packages i.e., Transformers.jl, PromptingTools.jl and ConformalPrediction.jl. Finally, I will conclude the talk by comparing the impact of large language models and small language models on the environment.
Outline:
1. Introduction to Large Language models and their impact on the environment(05 Minutes)
2. Overview of Small Language models as a solution to reduce the carbon footprint(05 Minutes)
3. Introduction to Transformers.jl, PromptingTools.jl and ConformalPrediction.jl (08 Minutes)
4. Building an example small language model application in Julia(08 Minutes)
5. Comparison of the environmental impact of large language models and small language models(02 Minutes)
6. Conclusion and Questions (02 Minutes)
Gajendra Deshpande is a distinguished professional with an M.Tech. in Computer Science and Engineering from Visvesvaraya Technological University, Belagavi, along with a PG Diploma in Cyber Law and Cyber Forensics from the National Law School of India University, Bengaluru. He founded and currently manages EyeSec Cyber Security Solutions Private Limited in Belagavi.
Deshpande is renowned for his extensive contribution to the tech community, having delivered over 100 talks and conducted more than 25 workshops at various esteemed international conferences, including JuliaCon 2023 at MIT, USA, EuroPython Ireland, PyCon MEA Dubai, PyCon APAC Japan, PyData Global, and many more across Europe, Asia and the USA. His expertise has guided teams to victory in the Smart India Hackathon and National Security Hackathon five times.
As an active member of PyCon India, Deshpande has played crucial roles, such as leading the Program Committee in 2021 and serving as the Mentorship Lead in 2023. He has been instrumental in organizing FOSSCon India 2019 and BelPy conferences. His commitment extends to various professional bodies, serving as the Vice Chair of the IEEE Young Professionals Affinity Group, Bangalore Section, and an Execom Member of IEEE NKSS. He is a Fellow Member of the Royal Statistical Society UK and maintains memberships with OWASP, the British Computer Society, and ACM. Deshpande has significantly contributed to Python, Julia, and FOSS Conferences by reviewing proposals, mentoring speakers, engaging in discussions, and organizing events.